Merge branch 'master' into embed_ui

.github/workflows/build.yml

@@ -70,7 +70,7 @@ jobs:
       - name: Get commit hash
         id: commit
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
-        uses: pr-mpt/actions-commit-hash@v2
+        uses: prompt/actions-commit-hash@v2
 
       - name: Fetch system info
         id: system-info
@@ -123,7 +123,7 @@ jobs:
       - name: Get commit hash
         id: commit
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
-        uses: pr-mpt/actions-commit-hash@v2
+        uses: prompt/actions-commit-hash@v2
 
       - name: Fetch system info
         id: system-info
@@ -162,7 +162,7 @@ jobs:
 
     strategy:
       matrix:
-        variant: [musa, sycl, vulkan]
+        variant: [musa, sycl, vulkan, cuda]
 
     env:
       REGISTRY: ghcr.io
@@ -177,7 +177,7 @@ jobs:
       - name: Get commit hash
         id: commit
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
-        uses: pr-mpt/actions-commit-hash@v2
+        uses: prompt/actions-commit-hash@v2
 
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
@@ -240,7 +240,7 @@ jobs:
       - name: Get commit hash
         id: commit
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
-        uses: pr-mpt/actions-commit-hash@v2
+        uses: prompt/actions-commit-hash@v2
 
       - name: Fetch system info
         id: system-info
@@ -340,7 +340,7 @@ jobs:
       - name: Get commit hash
         id: commit
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
-        uses: pr-mpt/actions-commit-hash@v2
+        uses: prompt/actions-commit-hash@v2
 
       - name: Pack artifacts
         id: pack_artifacts
@@ -463,7 +463,7 @@ jobs:
       - name: Get commit hash
         id: commit
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
-        uses: pr-mpt/actions-commit-hash@v2
+        uses: prompt/actions-commit-hash@v2
 
       - name: Pack artifacts
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
@@ -485,6 +485,146 @@ jobs:
           path: |
             sd-${{ env.BRANCH_NAME }}-${{ steps.commit.outputs.short }}-bin-win-rocm-x64.zip
 
+  ubuntu-latest-rocm:
+    runs-on: ubuntu-latest
+    container: rocm/dev-ubuntu-24.04:7.2
+
+    env:
+      ROCM_VERSION: "7.2"
+      UBUNTU_VERSION: "24.04"
+      GPU_TARGETS: "gfx1151;gfx1150;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+
+    steps:
+      - run: apt-get update && apt-get install -y git
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+        with:
+          submodules: recursive
+
+      - name: Free disk space
+        run: |
+          # Remove preinstalled SDKs and caches not needed for this job
+          sudo rm -rf /usr/share/dotnet || true
+          sudo rm -rf /usr/local/lib/android || true
+          sudo rm -rf /opt/ghc || true
+          sudo rm -rf /usr/local/.ghcup || true
+          sudo rm -rf /opt/hostedtoolcache || true
+
+          # Remove old package lists and caches
+          sudo rm -rf /var/lib/apt/lists/* || true
+          sudo apt clean
+
+      - name: Dependencies
+        id: depends
+        run: |
+          sudo apt-get update
+          sudo apt install -y \
+            cmake \
+            hip-dev \
+            hipblas-dev \
+            ninja-build \
+            rocm-dev \
+            zip
+          # Clean apt caches to recover disk space
+          sudo apt clean
+          sudo rm -rf /var/lib/apt/lists/* || true
+
+      - name: Setup ROCm Environment
+        run: |
+          # Add ROCm to PATH for current session
+          echo "/opt/rocm/bin" >> $GITHUB_PATH
+
+          # Build regex pattern from ${{ env.GPU_TARGETS }} (match target as substring)
+          TARGET_REGEX="($(printf '%s' "${{ env.GPU_TARGETS }}" | sed 's/;/|/g'))"
+
+          # Remove library files for architectures we're not building for to save disk space
+          echo "Cleaning up unneeded architecture files..."
+          cd /opt/rocm/lib/rocblas/library
+          # Keep only our target architectures
+          for file in *; do
+            if printf '%s' "$file" | grep -q 'gfx'; then
+              if ! printf '%s' "$file" | grep -Eq "$TARGET_REGEX"; then
+                echo "Removing $file" &&
+                sudo rm -f "$file";
+              fi
+            fi
+          done
+
+          cd /opt/rocm/lib/hipblaslt/library
+          for file in *; do
+            if printf '%s' "$file" | grep -q 'gfx'; then
+              if ! printf '%s' "$file" | grep -Eq "$TARGET_REGEX"; then
+                echo "Removing $file" &&
+                sudo rm -f "$file";
+              fi
+            fi
+          done
+
+      - name: Build
+        id: cmake_build
+        run: |
+          mkdir build
+          cd build
+          cmake .. -G Ninja \
+            -DCMAKE_CXX_COMPILER=amdclang++ \
+            -DCMAKE_C_COMPILER=amdclang \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DSD_HIPBLAS=ON \
+            -DGPU_TARGETS="${{ env.GPU_TARGETS }}" \
+            -DAMDGPU_TARGETS="${{ env.GPU_TARGETS }}" \
+            -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \
+            -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+            -DSD_BUILD_SHARED_LIBS=ON
+          cmake --build . --config Release
+
+      - name: Get commit hash
+        id: commit
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        uses: prompt/actions-commit-hash@v2
+
+      - name: Prepare artifacts
+        id: prepare_artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        run: |
+          # Copy licenses
+          cp ggml/LICENSE ./build/bin/ggml.txt
+          cp LICENSE ./build/bin/stable-diffusion.cpp.txt
+
+          # Move ROCm runtime libraries (to avoid double space consumption)
+          sudo mv /opt/rocm/lib/librocsparse.so* ./build/bin/
+          sudo mv /opt/rocm/lib/libhsa-runtime64.so* ./build/bin/
+          sudo mv /opt/rocm/lib/libamdhip64.so* ./build/bin/
+          sudo mv /opt/rocm/lib/libhipblas.so* ./build/bin/
+          sudo mv /opt/rocm/lib/libhipblaslt.so* ./build/bin/
+          sudo mv /opt/rocm/lib/librocblas.so* ./build/bin/
+          sudo mv /opt/rocm/lib/rocblas/ ./build/bin/
+          sudo mv /opt/rocm/lib/hipblaslt/ ./build/bin/
+
+      - name: Fetch system info
+        id: system-info
+        run: |
+          echo "CPU_ARCH=`uname -m`" >> "$GITHUB_OUTPUT"
+          echo "OS_NAME=`lsb_release -s -i`" >> "$GITHUB_OUTPUT"
+          echo "OS_VERSION=`lsb_release -s -r`" >> "$GITHUB_OUTPUT"
+          echo "OS_TYPE=`uname -s`" >> "$GITHUB_OUTPUT"
+
+      - name: Pack artifacts
+        id: pack_artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        run: |
+          cp ggml/LICENSE ./build/bin/ggml.txt
+          cp LICENSE ./build/bin/stable-diffusion.cpp.txt
+          zip -y -r sd-${{ env.BRANCH_NAME }}-${{ steps.commit.outputs.short }}-bin-${{ steps.system-info.outputs.OS_TYPE }}-Ubuntu-${{ env.UBUNTU_VERSION }}-${{ steps.system-info.outputs.CPU_ARCH }}-rocm.zip ./build/bin
+
+      - name: Upload artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: sd-${{ env.BRANCH_NAME }}-${{ steps.commit.outputs.short }}-bin-${{ steps.system-info.outputs.OS_TYPE }}-Ubuntu-${{ env.UBUNTU_VERSION }}-${{ steps.system-info.outputs.CPU_ARCH }}-rocm.zip
+          path: |
+            sd-${{ env.BRANCH_NAME }}-${{ steps.commit.outputs.short }}-bin-${{ steps.system-info.outputs.OS_TYPE }}-Ubuntu-${{ env.UBUNTU_VERSION }}-${{ steps.system-info.outputs.CPU_ARCH }}-rocm.zip
+
   release:
     if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
 
@@ -493,6 +633,7 @@ jobs:
     needs:
       - ubuntu-latest-cmake
       - ubuntu-latest-cmake-vulkan
+      - ubuntu-latest-rocm
       - build-and-push-docker-images
       - macOS-latest-cmake
       - windows-latest-cmake
@@ -519,7 +660,7 @@ jobs:
 
       - name: Get commit hash
         id: commit
-        uses: pr-mpt/actions-commit-hash@v2
+        uses: prompt/actions-commit-hash@v2
 
       - name: Create release
         id: create_release

CMakeLists.txt

@@ -36,7 +36,6 @@ option(SD_VULKAN                     "sd: vulkan backend" OFF)
 option(SD_OPENCL                     "sd: opencl backend" OFF)
 option(SD_SYCL                       "sd: sycl backend" OFF)
 option(SD_MUSA                       "sd: musa backend" OFF)
-option(SD_FAST_SOFTMAX               "sd: x1.5 faster softmax, indeterministic (sometimes, same seed don't generate same image), cuda only" OFF)
 option(SD_BUILD_SHARED_LIBS          "sd: build shared libs" OFF)
 option(SD_BUILD_SHARED_GGML_LIB      "sd: build ggml as a separate shared lib" OFF)
 option(SD_USE_SYSTEM_GGML            "sd: use system-installed GGML library" OFF)
@@ -70,26 +69,22 @@ if (SD_HIPBLAS)
     message("-- Use HIPBLAS as backend stable-diffusion")
     set(GGML_HIP ON)
     add_definitions(-DSD_USE_CUDA)
-    if(SD_FAST_SOFTMAX)
-        set(GGML_CUDA_FAST_SOFTMAX ON)
-    endif()
 endif ()
 
 if(SD_MUSA)
     message("-- Use MUSA as backend stable-diffusion")
     set(GGML_MUSA ON)
     add_definitions(-DSD_USE_CUDA)
-    if(SD_FAST_SOFTMAX)
-        set(GGML_CUDA_FAST_SOFTMAX ON)
-    endif()
 endif()
 
 set(SD_LIB stable-diffusion)
 
 file(GLOB SD_LIB_SOURCES
-    "*.h"
+    "src/*.h"
-    "*.cpp"
+    "src/*.cpp"
-    "*.hpp"
+    "src/*.hpp"
+    "src/vocab/*.h"
+    "src/vocab/*.cpp"
 )
 
 find_program(GIT_EXE NAMES git git.exe NO_CMAKE_FIND_ROOT_PATH)
@@ -119,7 +114,7 @@ endif()
 message(STATUS "stable-diffusion.cpp commit ${SDCPP_BUILD_COMMIT}")
 
 set_property(
-  SOURCE ${CMAKE_CURRENT_SOURCE_DIR}/version.cpp
+  SOURCE ${CMAKE_CURRENT_SOURCE_DIR}/src/version.cpp
   APPEND PROPERTY COMPILE_DEFINITIONS
   SDCPP_BUILD_COMMIT=${SDCPP_BUILD_COMMIT} SDCPP_BUILD_VERSION=${SDCPP_BUILD_VERSION}
 )
@@ -182,6 +177,7 @@ endif()
 add_subdirectory(thirdparty)
 
 target_link_libraries(${SD_LIB} PUBLIC ggml zip)
+target_include_directories(${SD_LIB} PUBLIC . include)
 target_include_directories(${SD_LIB} PUBLIC . thirdparty)
 target_compile_features(${SD_LIB} PUBLIC c_std_11 cxx_std_17)
 
@@ -190,7 +186,7 @@ if (SD_BUILD_EXAMPLES)
     add_subdirectory(examples)
 endif()
 
-set(SD_PUBLIC_HEADERS stable-diffusion.h)
+set(SD_PUBLIC_HEADERS include/stable-diffusion.h)
 set_target_properties(${SD_LIB} PROPERTIES PUBLIC_HEADER "${SD_PUBLIC_HEADERS}")
 
 install(TARGETS ${SD_LIB} LIBRARY PUBLIC_HEADER)

Dockerfile.cuda

@@ -0,0 +1,25 @@
+ARG CUDA_VERSION=12.6.3
+ARG UBUNTU_VERSION=24.04
+
+FROM nvidia/cuda:${CUDA_VERSION}-cudnn-devel-ubuntu${UBUNTU_VERSION} AS build
+
+RUN apt-get update && apt-get install -y --no-install-recommends build-essential git ccache cmake
+
+WORKDIR /sd.cpp
+
+COPY . .
+
+ARG CUDACXX=/usr/local/cuda/bin/nvcc
+RUN cmake . -B ./build -DSD_CUDA=ON
+RUN cmake --build ./build --config Release --parallel
+
+FROM nvidia/cuda:${CUDA_VERSION}-cudnn-runtime-ubuntu${UBUNTU_VERSION} AS runtime
+
+RUN apt-get update && \
+    apt-get install --yes --no-install-recommends libgomp1 && \
+    apt-get clean
+
+COPY --from=build /sd.cpp/build/bin/sd-cli /sd-cli
+COPY --from=build /sd.cpp/build/bin/sd-server /sd-server
+
+ENTRYPOINT [ "/sd-cli" ]

README.md

@@ -15,6 +15,9 @@ API and command-line option may change frequently.***
 
 ## 🔥Important News
 
+* **2026/01/18** 🚀 stable-diffusion.cpp now supports **FLUX.2-klein**  
+  👉 Details: [PR #1193](https://github.com/leejet/stable-diffusion.cpp/pull/1193)
+
 * **2025/12/01** 🚀 stable-diffusion.cpp now supports **Z-Image**  
   👉 Details: [PR #1020](https://github.com/leejet/stable-diffusion.cpp/pull/1020)
 
@@ -50,6 +53,7 @@ API and command-line option may change frequently.***
     - [Qwen Image](./docs/qwen_image.md)
     - [Z-Image](./docs/z_image.md)
     - [Ovis-Image](./docs/ovis_image.md)
+    - [Anima](./docs/anima.md)
   - Image Edit Models
     - [FLUX.1-Kontext-dev](./docs/kontext.md)
     - [Qwen Image Edit series](./docs/qwen_image_edit.md)
@@ -136,6 +140,7 @@ If you want to improve performance or reduce VRAM/RAM usage, please refer to [pe
 - [🔥Wan2.1/Wan2.2](./docs/wan.md)
 - [🔥Z-Image](./docs/z_image.md)
 - [Ovis-Image](./docs/ovis_image.md)
+- [Anima](./docs/anima.md)
 - [LoRA](./docs/lora.md)
 - [LCM/LCM-LoRA](./docs/lcm.md)
 - [Using PhotoMaker to personalize image generation](./docs/photo_maker.md)

docs/anima.md

@@ -0,0 +1,21 @@
+# How to Use
+
+## Download weights
+
+- Download Anima
+    - safetensors: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/diffusion_models
+    - gguf: https://huggingface.co/Bedovyy/Anima-GGUF/tree/main
+    - gguf Anima2: https://huggingface.co/JusteLeo/Anima2-GGUF/tree/main
+- Download vae
+    - safetensors: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/vae
+- Download Qwen3-0.6B-Base
+    - safetensors: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/text_encoders
+    - gguf: https://huggingface.co/mradermacher/Qwen3-0.6B-Base-GGUF/tree/main
+
+## Examples
+
+```sh
+.\bin\Release\sd-cli.exe --diffusion-model  ..\..\ComfyUI\models\diffusion_models\anima-preview.safetensors --vae ..\..\ComfyUI\models\vae\qwen_image_vae.safetensors  --llm ..\..\ComfyUI\models\text_encoders\qwen_3_06b_base.safetensors  -p "a lovely cat holding a sign says 'anima.cpp'" --cfg-scale 6.0 --sampling-method euler -v --offload-to-cpu --diffusion-fa
+```
+
+<img alt="anima image example" src="../assets/anima/example.png" />

docs/caching.md

@@ -11,6 +11,7 @@ Caching methods accelerate diffusion inference by reusing intermediate computati
 | `dbcache` | DiT models | Block-level L1 residual threshold |
 | `taylorseer` | DiT models | Taylor series approximation |
 | `cache-dit` | DiT models | Combined DBCache + TaylorSeer |
+| `spectrum` | UNET models | Chebyshev + Taylor output forecasting |
 
 ### UCache (UNET Models)
 
@@ -79,7 +80,7 @@ Uses Taylor series approximation to predict block outputs:
 Combines DBCache and TaylorSeer:
 
 ```bash
---cache-mode cache-dit --cache-preset fast
+--cache-mode cache-dit
 ```
 
 #### Parameters
@@ -91,14 +92,6 @@ Combines DBCache and TaylorSeer:
 | `threshold` | L1 residual difference threshold | 0.08 |
 | `warmup` | Steps before caching starts | 8 |
 
-#### Presets
-
-Available presets: `slow`, `medium`, `fast`, `ultra` (or `s`, `m`, `f`, `u`).
-
-```bash
---cache-mode cache-dit --cache-preset fast
-```
-
 #### SCM Options
 
 Steps Computation Mask controls which steps can be cached:
@@ -118,6 +111,28 @@ Mask values: `1` = compute, `0` = can cache.
 --scm-policy dynamic
 ```
 
+### Spectrum (UNET Models)
+
+Spectrum uses Chebyshev polynomial fitting blended with Taylor extrapolation to predict denoised outputs, skipping entire UNet forward passes. Based on the paper [Spectrum: Adaptive Spectral Feature Forecasting for Efficient Diffusion Sampling](https://github.com/tingyu215/Spectrum).
+
+```bash
+sd-cli -m model.safetensors -p "a cat" --cache-mode spectrum
+```
+
+#### Parameters
+
+| Parameter | Description | Default |
+|-----------|-------------|---------|
+| `w` | Chebyshev vs Taylor blend weight (0=Taylor, 1=Chebyshev) | 0.40 |
+| `m` | Chebyshev polynomial degree | 3 |
+| `lam` | Ridge regression regularization | 1.0 |
+| `window` | Initial window size (compute every N steps) | 2 |
+| `flex` | Window growth per computed step after warmup | 0.50 |
+| `warmup` | Steps to always compute before caching starts | 4 |
+| `stop` | Stop caching at this fraction of total steps | 0.9 |
+
+```
+
 ### Performance Tips
 
 - Start with default thresholds and adjust based on output quality

docs/distilled_sd.md

@@ -1,8 +1,8 @@
-# Running distilled models: SSD1B and SDx.x with tiny U-Nets
+# Running distilled models: SSD1B, Vega and SDx.x with tiny U-Nets
 
 ## Preface 
 
-These models feature a reduced U-Net architecture. Unlike standard SDXL models, the SSD-1B U-Net contains only one middle block and fewer attention layers in its up- and down-blocks, resulting in significantly smaller file sizes. Using these models can reduce inference time by more than 33%. For more details, refer to Segmind's paper: https://arxiv.org/abs/2401.02677v1.
+These models feature a reduced U-Net architecture. Unlike standard SDXL models, the SSD-1B and Vega U-Net contains only one middle block and fewer attention layers in its up- and down-blocks, resulting in significantly smaller file sizes. Using these models can reduce inference time by more than 33%. For more details, refer to Segmind's paper: https://arxiv.org/abs/2401.02677v1.
 Similarly, SD1.x- and SD2.x-style models with a tiny U-Net consist of only 6 U-Net blocks, leading to very small files and time savings of up to 50%. For more information, see the paper: https://arxiv.org/pdf/2305.15798.pdf.
 
 ## SSD1B
@@ -17,7 +17,17 @@ Useful LoRAs are also available:
  * https://huggingface.co/seungminh/lora-swarovski-SSD-1B/resolve/main/pytorch_lora_weights.safetensors
  * https://huggingface.co/kylielee505/mylcmlorassd/resolve/main/pytorch_lora_weights.safetensors
 
-These files can be used out-of-the-box, unlike the models described in the next section.
+## Vega
+
+Segmind's Vega model is available online here:
+
+ * https://huggingface.co/segmind/Segmind-Vega/resolve/main/segmind-vega.safetensors
+ 
+VegaRT is an example for an LCM-LoRA:
+
+ * https://huggingface.co/segmind/Segmind-VegaRT/resolve/main/pytorch_lora_weights.safetensors
+
+Both files can be used out-of-the-box, unlike the models described in next sections.
 
 
 ## SD1.x, SD2.x with tiny U-Nets

docs/esrgan.md

@@ -1,6 +1,6 @@
 ## Using ESRGAN to upscale results
 
-You can use ESRGAN to upscale the generated images. At the moment, only the [RealESRGAN_x4plus_anime_6B.pth](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth) model is supported. Support for more models of this architecture will be added soon.
+You can use ESRGAN—such as the model [RealESRGAN_x4plus_anime_6B.pth](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth)—to upscale the generated images and improve their overall resolution and clarity.
 
 - Specify the model path using the `--upscale-model PATH` parameter. example:
 

docs/z_image.md

@@ -7,6 +7,9 @@ You can run Z-Image with stable-diffusion.cpp on GPUs with 4GB of VRAM — or ev
 - Download Z-Image-Turbo
     - safetensors: https://huggingface.co/Comfy-Org/z_image_turbo/tree/main/split_files/diffusion_models
     - gguf: https://huggingface.co/leejet/Z-Image-Turbo-GGUF/tree/main
+- Download Z-Image
+    - safetensors: https://huggingface.co/Comfy-Org/z_image/tree/main/split_files/diffusion_models
+    - gguf: https://huggingface.co/unsloth/Z-Image-GGUF/tree/main
 - Download vae
     - safetensors: https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main
 - Download Qwen3 4b
@@ -15,12 +18,22 @@ You can run Z-Image with stable-diffusion.cpp on GPUs with 4GB of VRAM — or ev
 
 ## Examples
 
+### Z-Image-Turbo
+
 ```
 .\bin\Release\sd-cli.exe --diffusion-model  z_image_turbo-Q3_K.gguf --vae ..\..\ComfyUI\models\vae\ae.sft  --llm ..\..\ComfyUI\models\text_encoders\Qwen3-4B-Instruct-2507-Q4_K_M.gguf -p "A cinematic, melancholic photograph of a solitary hooded figure walking through a sprawling, rain-slicked metropolis at night. The city lights are a chaotic blur of neon orange and cool blue, reflecting on the wet asphalt. The scene evokes a sense of being a single component in a vast machine. Superimposed over the image in a sleek, modern, slightly glitched font is the philosophical quote: 'THE CITY IS A CIRCUIT BOARD, AND I AM A BROKEN TRANSISTOR.' -- moody, atmospheric, profound, dark academic" --cfg-scale 1.0 -v --offload-to-cpu --diffusion-fa -H 1024 -W 512
 ```
 
 <img width="256" alt="z-image example" src="../assets/z_image/q3_K.png" />
 
+### Z-Image-Base
+
+```
+.\bin\Release\sd-cli.exe --diffusion-model  ..\..\ComfyUI\models\diffusion_models\z_image_bf16.safetensors --vae ..\..\ComfyUI\models\vae\ae.sft  --llm ..\..\ComfyUI\models\text_encoders\qwen_3_4b.safetensors -p "A cinematic, melancholic photograph of a solitary hooded figure walking through a sprawling, rain-slicked metropolis at night. The city lights are a chaotic blur of neon orange and cool blue, reflecting on the wet asphalt. The scene evokes a sense of being a single component in a vast machine. Superimposed over the image in a sleek, modern, slightly glitched font is the philosophical quote: 'THE CITY IS A CIRCUIT BOARD, AND I AM A BROKEN TRANSISTOR.' -- moody, atmospheric, profound, dark academic" --cfg-scale 5.0 -v --offload-to-cpu --diffusion-fa -H 1024 -W 512
+```
+
+<img width="256" alt="z-image example" src="../assets/z_image/base_bf16.png" />
+
 ## Comparison of Different Quantization Types
 
 | bf16 | q8_0 | q6_K | q5_0 | q4_K | q4_0 | q3_K | q2_K|

examples/cli/README.md

@@ -4,11 +4,12 @@
 usage: ./bin/sd-cli  [options]
 
 CLI Options:
-  -o, --output <string>       path to write result image to. you can use printf-style %d format specifiers for image sequences (default: ./output.png) (eg. output_%03d.png)
+  -o, --output <string>       path to write result image to. you can use printf-style %d format specifiers for image sequences (default:
-  --output-begin-idx <int>    starting index for output image sequence, must be non-negative (default 0 if specified %d in output path, 1 otherwise)
+                              ./output.png) (eg. output_%03d.png)
   --preview-path <string>     path to write preview image to (default: ./preview.png)
   --preview-interval <int>    interval in denoising steps between consecutive updates of the image preview file (default is 1, meaning updating at
                               every step)
+  --output-begin-idx <int>    starting index for output image sequence, must be non-negative (default 0 if specified %d in output path, 1 otherwise)
   --canny                     apply canny preprocessor (edge detection)
   --convert-name              convert tensor name (for convert mode)
   -v, --verbose               print extra info
@@ -44,7 +45,6 @@ Context Options:
                                            CPU physical cores
   --chroma-t5-mask-pad <int>               t5 mask pad size of chroma
   --vae-tile-overlap <float>               tile overlap for vae tiling, in fraction of tile size (default: 0.5)
-  --flow-shift <float>                     shift value for Flow models like SD3.x or WAN (default: auto)
   --vae-tiling                             process vae in tiles to reduce memory usage
   --force-sdxl-vae-conv-scale              force use of conv scale on sdxl vae
   --offload-to-cpu                         place the weights in RAM to save VRAM, and automatically load them into VRAM when needed
@@ -52,13 +52,15 @@ Context Options:
   --control-net-cpu                        keep controlnet in cpu (for low vram)
   --clip-on-cpu                            keep clip in cpu (for low vram)
   --vae-on-cpu                             keep vae in cpu (for low vram)
-  --diffusion-fa                           use flash attention in the diffusion model
+  --fa                                     use flash attention
+  --diffusion-fa                           use flash attention in the diffusion model only
   --diffusion-conv-direct                  use ggml_conv2d_direct in the diffusion model
   --vae-conv-direct                        use ggml_conv2d_direct in the vae model
   --circular                               enable circular padding for convolutions
   --circularx                              enable circular RoPE wrapping on x-axis (width) only
   --circulary                              enable circular RoPE wrapping on y-axis (height) only
   --chroma-disable-dit-mask                disable dit mask for chroma
+  --qwen-image-zero-cond-t                 enable zero_cond_t for qwen image
   --chroma-enable-t5-mask                  enable t5 mask for chroma
   --type                                   weight type (examples: f32, f16, q4_0, q4_1, q5_0, q5_1, q8_0, q2_K, q3_K, q4_K). If not specified, the default is the
                                            type of the weight file
@@ -108,6 +110,7 @@ Generation Options:
   --skip-layer-start <float>               SLG enabling point (default: 0.01)
   --skip-layer-end <float>                 SLG disabling point (default: 0.2)
   --eta <float>                            eta in DDIM, only for DDIM and TCD (default: 0)
+  --flow-shift <float>                     shift value for Flow models like SD3.x or WAN (default: auto)
   --high-noise-cfg-scale <float>           (high noise) unconditional guidance scale: (default: 7.0)
   --high-noise-img-cfg-scale <float>       (high noise) image guidance scale for inpaint or instruct-pix2pix models (default: same as --cfg-scale)
   --high-noise-guidance <float>            (high noise) distilled guidance scale for models with guidance input (default: 3.5)
@@ -124,20 +127,23 @@ Generation Options:
   --disable-auto-resize-ref-image          disable auto resize of ref images
   -s, --seed                               RNG seed (default: 42, use random seed for < 0)
   --sampling-method                        sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing,
-                                           tcd] (default: euler for Flux/SD3/Wan, euler_a otherwise)
+                                           tcd, res_multistep, res_2s] (default: euler for Flux/SD3/Wan, euler_a
+                                           otherwise)
   --high-noise-sampling-method             (high noise) sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm,
-                                           ddim_trailing, tcd] default: euler for Flux/SD3/Wan, euler_a otherwise
+                                           ddim_trailing, tcd, res_multistep, res_2s] default: euler for Flux/SD3/Wan,
+                                           euler_a otherwise
   --scheduler                              denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits, smoothstep, sgm_uniform, simple,
-                                           kl_optimal, lcm], default: discrete
+                                           kl_optimal, lcm, bong_tangent], default: discrete
   --sigmas                                 custom sigma values for the sampler, comma-separated (e.g., "14.61,7.8,3.5,0.0").
   --skip-layers                            layers to skip for SLG steps (default: [7,8,9])
   --high-noise-skip-layers                 (high noise) layers to skip for SLG steps (default: [7,8,9])
   -r, --ref-image                          reference image for Flux Kontext models (can be used multiple times)
-  --cache-mode                             caching method: 'easycache' (DiT), 'ucache' (UNET), 'dbcache'/'taylorseer'/'cache-dit' (DiT block-level)
+  --cache-mode                             caching method: 'easycache' (DiT), 'ucache' (UNET), 'dbcache'/'taylorseer'/'cache-dit' (DiT block-level),
+                                           'spectrum' (UNET/DiT Chebyshev+Taylor forecasting)
   --cache-option                           named cache params (key=value format, comma-separated). easycache/ucache:
-                                           threshold=,start=,end=,decay=,relative=,reset=; dbcache/taylorseer/cache-dit: Fn=,Bn=,threshold=,warmup=. Examples:
+                                           threshold=,start=,end=,decay=,relative=,reset=; dbcache/taylorseer/cache-dit: Fn=,Bn=,threshold=,warmup=;
-                                           "threshold=0.25" or "threshold=1.5,reset=0"
+                                           spectrum: w=,m=,lam=,window=,flex=,warmup=,stop=. Examples:
-  --cache-preset                           cache-dit preset: 'slow'/'s', 'medium'/'m', 'fast'/'f', 'ultra'/'u'
+                                           "threshold=0.25" or "threshold=1.5,reset=0" or "w=0.4,window=2"
   --scm-mask                               SCM steps mask for cache-dit: comma-separated 0/1 (e.g., "1,1,1,0,0,1,0,0,1,0") - 1=compute, 0=can cache
   --scm-policy                             SCM policy: 'dynamic' (default) or 'static'
 ```

examples/cli/main.cpp

@@ -245,7 +245,7 @@ std::string get_image_params(const SDCliParams& cli_params, const SDContextParam
     parameter_string += "Guidance: " + std::to_string(gen_params.sample_params.guidance.distilled_guidance) + ", ";
     parameter_string += "Eta: " + std::to_string(gen_params.sample_params.eta) + ", ";
     parameter_string += "Seed: " + std::to_string(seed) + ", ";
-    parameter_string += "Size: " + std::to_string(gen_params.width) + "x" + std::to_string(gen_params.height) + ", ";
+    parameter_string += "Size: " + std::to_string(gen_params.get_resolved_width()) + "x" + std::to_string(gen_params.get_resolved_height()) + ", ";
     parameter_string += "Model: " + sd_basename(ctx_params.model_path) + ", ";
     parameter_string += "RNG: " + std::string(sd_rng_type_name(ctx_params.rng_type)) + ", ";
     if (ctx_params.sampler_rng_type != RNG_TYPE_COUNT) {
@@ -394,12 +394,15 @@ bool save_results(const SDCliParams& cli_params,
 
     fs::path base_path = out_path;
     fs::path ext       = out_path.has_extension() ? out_path.extension() : fs::path{};
-    if (!ext.empty())
-        base_path.replace_extension();
 
     std::string ext_lower = ext.string();
     std::transform(ext_lower.begin(), ext_lower.end(), ext_lower.begin(), ::tolower);
     bool is_jpg = (ext_lower == ".jpg" || ext_lower == ".jpeg" || ext_lower == ".jpe");
+    if (!ext.empty()) {
+        if (is_jpg || ext_lower == ".png") {
+            base_path.replace_extension();
+        }
+    }
 
     int output_begin_idx = cli_params.output_begin_idx;
     if (output_begin_idx < 0) {
@@ -409,7 +412,7 @@ bool save_results(const SDCliParams& cli_params,
     auto write_image = [&](const fs::path& path, int idx) {
         const sd_image_t& img = results[idx];
         if (!img.data)
-            return;
+            return false;
 
         std::string params = get_image_params(cli_params, ctx_params, gen_params, gen_params.seed + idx);
         int ok             = 0;
@@ -419,8 +422,11 @@ bool save_results(const SDCliParams& cli_params,
             ok = stbi_write_png(path.string().c_str(), img.width, img.height, img.channel, img.data, 0, params.c_str());
         }
         LOG_INFO("save result image %d to '%s' (%s)", idx, path.string().c_str(), ok ? "success" : "failure");
+        return ok != 0;
     };
 
+    int sucessful_reults = 0;
+
     if (std::regex_search(cli_params.output_path, format_specifier_regex)) {
         if (!is_jpg && ext_lower != ".png")
             ext = ".png";
@@ -429,9 +435,12 @@ bool save_results(const SDCliParams& cli_params,
 
         for (int i = 0; i < num_results; ++i) {
             fs::path img_path = format_frame_idx(pattern.string(), output_begin_idx + i);
-            write_image(img_path, i);
+            if (write_image(img_path, i)) {
+                sucessful_reults++;
             }
-        return true;
+        }
+        LOG_INFO("%d/%d images saved", sucessful_reults, num_results);
+        return sucessful_reults != 0;
     }
 
     if (cli_params.mode == VID_GEN && num_results > 1) {
@@ -439,9 +448,13 @@ bool save_results(const SDCliParams& cli_params,
             ext = ".avi";
         fs::path video_path = base_path;
         video_path += ext;
-        create_mjpg_avi_from_sd_images(video_path.string().c_str(), results, num_results, gen_params.fps);
+        if (create_mjpg_avi_from_sd_images(video_path.string().c_str(), results, num_results, gen_params.fps) == 0) {
             LOG_INFO("save result MJPG AVI video to '%s'", video_path.string().c_str());
             return true;
+        } else {
+            LOG_ERROR("Failed to save result MPG AVI video to '%s'", video_path.string().c_str());
+            return false;
+        }
     }
 
     if (!is_jpg && ext_lower != ".png")
@@ -453,10 +466,12 @@ bool save_results(const SDCliParams& cli_params,
             img_path += "_" + std::to_string(output_begin_idx + i);
         }
         img_path += ext;
-        write_image(img_path, i);
+        if (write_image(img_path, i)) {
+            sucessful_reults++;
         }
-
+    }
-    return true;
+    LOG_INFO("%d/%d images saved", sucessful_reults, num_results);
+    return sucessful_reults != 0;
 }
 
 int main(int argc, const char* argv[]) {
@@ -526,10 +541,10 @@ int main(int argc, const char* argv[]) {
     }
 
     bool vae_decode_only     = true;
-    sd_image_t init_image    = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 3, nullptr};
+    sd_image_t init_image    = {0, 0, 3, nullptr};
-    sd_image_t end_image     = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 3, nullptr};
+    sd_image_t end_image     = {0, 0, 3, nullptr};
-    sd_image_t control_image = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 3, nullptr};
+    sd_image_t control_image = {0, 0, 3, nullptr};
-    sd_image_t mask_image    = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 1, nullptr};
+    sd_image_t mask_image    = {0, 0, 1, nullptr};
     std::vector<sd_image_t> ref_images;
     std::vector<sd_image_t> pmid_images;
     std::vector<sd_image_t> control_frames;
@@ -556,57 +571,79 @@ int main(int argc, const char* argv[]) {
         control_frames.clear();
     };
 
+    auto load_image_and_update_size = [&](const std::string& path,
+                                          sd_image_t& image,
+                                          bool resize_image    = true,
+                                          int expected_channel = 3) -> bool {
+        int expected_width  = 0;
+        int expected_height = 0;
+        if (resize_image && gen_params.width_and_height_are_set()) {
+            expected_width  = gen_params.width;
+            expected_height = gen_params.height;
+        }
+
+        if (!load_sd_image_from_file(&image, path.c_str(), expected_width, expected_height, expected_channel)) {
+            LOG_ERROR("load image from '%s' failed", path.c_str());
+            release_all_resources();
+            return false;
+        }
+
+        gen_params.set_width_and_height_if_unset(image.width, image.height);
+        return true;
+    };
+
     if (gen_params.init_image_path.size() > 0) {
         vae_decode_only = false;
-
+        if (!load_image_and_update_size(gen_params.init_image_path, init_image)) {
-        int width       = 0;
-        int height      = 0;
-        init_image.data = load_image_from_file(gen_params.init_image_path.c_str(), width, height, gen_params.width, gen_params.height);
-        if (init_image.data == nullptr) {
-            LOG_ERROR("load image from '%s' failed", gen_params.init_image_path.c_str());
-            release_all_resources();
             return 1;
         }
     }
 
     if (gen_params.end_image_path.size() > 0) {
         vae_decode_only = false;
-
+        if (!load_image_and_update_size(gen_params.init_image_path, end_image)) {
-        int width      = 0;
-        int height     = 0;
-        end_image.data = load_image_from_file(gen_params.end_image_path.c_str(), width, height, gen_params.width, gen_params.height);
-        if (end_image.data == nullptr) {
-            LOG_ERROR("load image from '%s' failed", gen_params.end_image_path.c_str());
-            release_all_resources();
             return 1;
         }
     }
 
+    if (gen_params.ref_image_paths.size() > 0) {
+        vae_decode_only = false;
+        for (auto& path : gen_params.ref_image_paths) {
+            sd_image_t ref_image = {0, 0, 3, nullptr};
+            if (!load_image_and_update_size(path, ref_image, false)) {
+                return 1;
+            }
+            ref_images.push_back(ref_image);
+        }
+    }
+
     if (gen_params.mask_image_path.size() > 0) {
-        int c           = 0;
+        if (!load_sd_image_from_file(&mask_image,
-        int width       = 0;
+                                     gen_params.mask_image_path.c_str(),
-        int height      = 0;
+                                     gen_params.get_resolved_width(),
-        mask_image.data = load_image_from_file(gen_params.mask_image_path.c_str(), width, height, gen_params.width, gen_params.height, 1);
+                                     gen_params.get_resolved_height(),
-        if (mask_image.data == nullptr) {
+                                     1)) {
             LOG_ERROR("load image from '%s' failed", gen_params.mask_image_path.c_str());
             release_all_resources();
             return 1;
         }
     } else {
-        mask_image.data = (uint8_t*)malloc(gen_params.width * gen_params.height);
+        mask_image.data = (uint8_t*)malloc(gen_params.get_resolved_width() * gen_params.get_resolved_height());
         if (mask_image.data == nullptr) {
             LOG_ERROR("malloc mask image failed");
             release_all_resources();
             return 1;
         }
-        memset(mask_image.data, 255, gen_params.width * gen_params.height);
+        mask_image.width  = gen_params.get_resolved_width();
+        mask_image.height = gen_params.get_resolved_height();
+        memset(mask_image.data, 255, gen_params.get_resolved_width() * gen_params.get_resolved_height());
     }
 
     if (gen_params.control_image_path.size() > 0) {
-        int width          = 0;
+        if (!load_sd_image_from_file(&control_image,
-        int height         = 0;
+                                     gen_params.control_image_path.c_str(),
-        control_image.data = load_image_from_file(gen_params.control_image_path.c_str(), width, height, gen_params.width, gen_params.height);
+                                     gen_params.get_resolved_width(),
-        if (control_image.data == nullptr) {
+                                     gen_params.get_resolved_height())) {
             LOG_ERROR("load image from '%s' failed", gen_params.control_image_path.c_str());
             release_all_resources();
             return 1;
@@ -621,29 +658,11 @@ int main(int argc, const char* argv[]) {
         }
     }
 
-    if (gen_params.ref_image_paths.size() > 0) {
-        vae_decode_only = false;
-        for (auto& path : gen_params.ref_image_paths) {
-            int width             = 0;
-            int height            = 0;
-            uint8_t* image_buffer = load_image_from_file(path.c_str(), width, height);
-            if (image_buffer == nullptr) {
-                LOG_ERROR("load image from '%s' failed", path.c_str());
-                release_all_resources();
-                return 1;
-            }
-            ref_images.push_back({(uint32_t)width,
-                                  (uint32_t)height,
-                                  3,
-                                  image_buffer});
-        }
-    }
-
     if (!gen_params.control_video_path.empty()) {
         if (!load_images_from_dir(gen_params.control_video_path,
                                   control_frames,
-                                  gen_params.width,
+                                  gen_params.get_resolved_width(),
-                                  gen_params.height,
+                                  gen_params.get_resolved_height(),
                                   gen_params.video_frames,
                                   cli_params.verbose)) {
             release_all_resources();
@@ -717,8 +736,8 @@ int main(int argc, const char* argv[]) {
                 gen_params.auto_resize_ref_image,
                 gen_params.increase_ref_index,
                 mask_image,
-                gen_params.width,
+                gen_params.get_resolved_width(),
-                gen_params.height,
+                gen_params.get_resolved_height(),
                 gen_params.sample_params,
                 gen_params.strength,
                 gen_params.seed,
@@ -748,8 +767,8 @@ int main(int argc, const char* argv[]) {
                 end_image,
                 control_frames.data(),
                 (int)control_frames.size(),
-                gen_params.width,
+                gen_params.get_resolved_width(),
-                gen_params.height,
+                gen_params.get_resolved_height(),
                 gen_params.sample_params,
                 gen_params.high_noise_sample_params,
                 gen_params.moe_boundary,

examples/common/common.hpp

@@ -445,7 +445,7 @@ struct SDContextParams {
     std::string photo_maker_path;
     sd_type_t wtype = SD_TYPE_COUNT;
     std::string tensor_type_rules;
-    std::string lora_model_dir;
+    std::string lora_model_dir = ".";
 
     std::map<std::string, std::string> embedding_map;
     std::vector<sd_embedding_t> embedding_vec;
@@ -457,6 +457,7 @@ struct SDContextParams {
     bool control_net_cpu        = false;
     bool clip_on_cpu            = false;
     bool vae_on_cpu             = false;
+    bool flash_attn             = false;
     bool diffusion_flash_attn   = false;
     bool diffusion_conv_direct  = false;
     bool vae_conv_direct        = false;
@@ -580,10 +581,6 @@ struct SDContextParams {
              "--vae-tile-overlap",
              "tile overlap for vae tiling, in fraction of tile size (default: 0.5)",
              &vae_tiling_params.target_overlap},
-            {"",
-             "--flow-shift",
-             "shift value for Flow models like SD3.x or WAN (default: auto)",
-             &flow_shift},
         };
 
         options.bool_options = {
@@ -615,9 +612,13 @@ struct SDContextParams {
              "--vae-on-cpu",
              "keep vae in cpu (for low vram)",
              true, &vae_on_cpu},
+            {"",
+             "--fa",
+             "use flash attention",
+             true, &flash_attn},
             {"",
              "--diffusion-fa",
-             "use flash attention in the diffusion model",
+             "use flash attention in the diffusion model only",
              true, &diffusion_flash_attn},
             {"",
              "--diffusion-conv-direct",
@@ -898,12 +899,12 @@ struct SDContextParams {
             << "  photo_maker_path: \"" << photo_maker_path << "\",\n"
             << "  rng_type: " << sd_rng_type_name(rng_type) << ",\n"
             << "  sampler_rng_type: " << sd_rng_type_name(sampler_rng_type) << ",\n"
-            << "  flow_shift: " << (std::isinf(flow_shift) ? "INF" : std::to_string(flow_shift)) << "\n"
             << "  offload_params_to_cpu: " << (offload_params_to_cpu ? "true" : "false") << ",\n"
             << "  enable_mmap: " << (enable_mmap ? "true" : "false") << ",\n"
             << "  control_net_cpu: " << (control_net_cpu ? "true" : "false") << ",\n"
             << "  clip_on_cpu: " << (clip_on_cpu ? "true" : "false") << ",\n"
             << "  vae_on_cpu: " << (vae_on_cpu ? "true" : "false") << ",\n"
+            << "  flash_attn: " << (flash_attn ? "true" : "false") << ",\n"
             << "  diffusion_flash_attn: " << (diffusion_flash_attn ? "true" : "false") << ",\n"
             << "  diffusion_conv_direct: " << (diffusion_conv_direct ? "true" : "false") << ",\n"
             << "  vae_conv_direct: " << (vae_conv_direct ? "true" : "false") << ",\n"
@@ -968,6 +969,7 @@ struct SDContextParams {
             clip_on_cpu,
             control_net_cpu,
             vae_on_cpu,
+            flash_attn,
             diffusion_flash_attn,
             taesd_preview,
             diffusion_conv_direct,
@@ -979,7 +981,6 @@ struct SDContextParams {
             chroma_use_t5_mask,
             chroma_t5_mask_pad,
             qwen_image_zero_cond_t,
-            flow_shift,
         };
         return sd_ctx_params;
     }
@@ -1024,8 +1025,8 @@ struct SDGenerationParams {
     std::string prompt_with_lora;  // for metadata record only
     std::string negative_prompt;
     int clip_skip   = -1;  // <= 0 represents unspecified
-    int width       = 512;
+    int width       = -1;
-    int height      = 512;
+    int height      = -1;
     int batch_count = 1;
     std::string init_image_path;
     std::string end_image_path;
@@ -1046,7 +1047,6 @@ struct SDGenerationParams {
 
     std::string cache_mode;
     std::string cache_option;
-    std::string cache_preset;
     std::string scm_mask;
     bool scm_policy_dynamic = true;
     sd_cache_params_t cache_params{};
@@ -1199,6 +1199,10 @@ struct SDGenerationParams {
              "--eta",
              "eta in DDIM, only for DDIM and TCD (default: 0)",
              &sample_params.eta},
+            {"",
+             "--flow-shift",
+             "shift value for Flow models like SD3.x or WAN (default: auto)",
+             &sample_params.flow_shift},
             {"",
              "--high-noise-cfg-scale",
              "(high noise) unconditional guidance scale: (default: 7.0)",
@@ -1417,8 +1421,8 @@ struct SDGenerationParams {
             }
             cache_mode = argv_to_utf8(index, argv);
             if (cache_mode != "easycache" && cache_mode != "ucache" &&
-                cache_mode != "dbcache" && cache_mode != "taylorseer" && cache_mode != "cache-dit") {
+                cache_mode != "dbcache" && cache_mode != "taylorseer" && cache_mode != "cache-dit" && cache_mode != "spectrum") {
-                fprintf(stderr, "error: invalid cache mode '%s', must be 'easycache', 'ucache', 'dbcache', 'taylorseer', or 'cache-dit'\n", cache_mode.c_str());
+                fprintf(stderr, "error: invalid cache mode '%s', must be 'easycache', 'ucache', 'dbcache', 'taylorseer', 'cache-dit', or 'spectrum'\n", cache_mode.c_str());
                 return -1;
             }
             return 1;
@@ -1456,21 +1460,6 @@ struct SDGenerationParams {
             return 1;
         };
 
-        auto on_cache_preset_arg = [&](int argc, const char** argv, int index) {
-            if (++index >= argc) {
-                return -1;
-            }
-            cache_preset = argv_to_utf8(index, argv);
-            if (cache_preset != "slow" && cache_preset != "s" && cache_preset != "S" &&
-                cache_preset != "medium" && cache_preset != "m" && cache_preset != "M" &&
-                cache_preset != "fast" && cache_preset != "f" && cache_preset != "F" &&
-                cache_preset != "ultra" && cache_preset != "u" && cache_preset != "U") {
-                fprintf(stderr, "error: invalid cache preset '%s', must be 'slow'/'s', 'medium'/'m', 'fast'/'f', or 'ultra'/'u'\n", cache_preset.c_str());
-                return -1;
-            }
-            return 1;
-        };
-
         options.manual_options = {
             {"-s",
              "--seed",
@@ -1478,17 +1467,17 @@ struct SDGenerationParams {
              on_seed_arg},
             {"",
              "--sampling-method",
-             "sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd] "
+             "sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd, res_multistep, res_2s] "
              "(default: euler for Flux/SD3/Wan, euler_a otherwise)",
              on_sample_method_arg},
             {"",
              "--high-noise-sampling-method",
-             "(high noise) sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd]"
+             "(high noise) sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd, res_multistep, res_2s]"
              " default: euler for Flux/SD3/Wan, euler_a otherwise",
              on_high_noise_sample_method_arg},
             {"",
              "--scheduler",
-             "denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits, smoothstep, sgm_uniform, simple, kl_optimal, lcm], default: discrete",
+             "denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits, smoothstep, sgm_uniform, simple, kl_optimal, lcm, bong_tangent], default: discrete",
              on_scheduler_arg},
             {"",
              "--sigmas",
@@ -1508,16 +1497,12 @@ struct SDGenerationParams {
              on_ref_image_arg},
             {"",
              "--cache-mode",
-             "caching method: 'easycache' (DiT), 'ucache' (UNET), 'dbcache'/'taylorseer'/'cache-dit' (DiT block-level)",
+             "caching method: 'easycache' (DiT), 'ucache' (UNET), 'dbcache'/'taylorseer'/'cache-dit' (DiT block-level), 'spectrum' (UNET/DiT Chebyshev+Taylor forecasting)",
              on_cache_mode_arg},
             {"",
              "--cache-option",
-             "named cache params (key=value format, comma-separated). easycache/ucache: threshold=,start=,end=,decay=,relative=,reset=; dbcache/taylorseer/cache-dit: Fn=,Bn=,threshold=,warmup=. Examples: \"threshold=0.25\" or \"threshold=1.5,reset=0\"",
+             "named cache params (key=value format, comma-separated). easycache/ucache: threshold=,start=,end=,decay=,relative=,reset=; dbcache/taylorseer/cache-dit: Fn=,Bn=,threshold=,warmup=; spectrum: w=,m=,lam=,window=,flex=,warmup=,stop=. Examples: \"threshold=0.25\" or \"threshold=1.5,reset=0\"",
              on_cache_option_arg},
-            {"",
-             "--cache-preset",
-             "cache-dit preset: 'slow'/'s', 'medium'/'m', 'fast'/'f', 'ultra'/'u'",
-             on_cache_preset_arg},
             {"",
              "--scm-mask",
              "SCM steps mask for cache-dit: comma-separated 0/1 (e.g., \"1,1,1,0,0,1,0,0,1,0\") - 1=compute, 0=can cache",
@@ -1570,7 +1555,6 @@ struct SDGenerationParams {
         load_if_exists("negative_prompt", negative_prompt);
         load_if_exists("cache_mode", cache_mode);
         load_if_exists("cache_option", cache_option);
-        load_if_exists("cache_preset", cache_preset);
         load_if_exists("scm_mask", scm_mask);
 
         load_if_exists("clip_skip", clip_skip);
@@ -1599,6 +1583,7 @@ struct SDGenerationParams {
         load_if_exists("cfg_scale", sample_params.guidance.txt_cfg);
         load_if_exists("img_cfg_scale", sample_params.guidance.img_cfg);
         load_if_exists("guidance", sample_params.guidance.distilled_guidance);
+        load_if_exists("flow_shift", sample_params.flow_shift);
 
         auto load_sampler_if_exists = [&](const char* key, enum sample_method_t& out) {
             if (j.contains(key) && j[key].is_string()) {
@@ -1705,17 +1690,24 @@ struct SDGenerationParams {
         }
     }
 
-    bool process_and_check(SDMode mode, const std::string& lora_model_dir) {
+    bool width_and_height_are_set() const {
-        prompt_with_lora = prompt;
+        return width > 0 && height > 0;
-        if (width <= 0) {
-            LOG_ERROR("error: the width must be greater than 0\n");
-            return false;
     }
 
-        if (height <= 0) {
+    void set_width_and_height_if_unset(int w, int h) {
-            LOG_ERROR("error: the height must be greater than 0\n");
+        if (!width_and_height_are_set()) {
-            return false;
+            LOG_INFO("set width x height to %d x %d", w, h);
+            width  = w;
+            height = h;
         }
+    }
+
+    int get_resolved_width() const { return (width > 0) ? width : 512; }
+
+    int get_resolved_height() const { return (height > 0) ? height : 512; }
+
+    bool process_and_check(SDMode mode, const std::string& lora_model_dir) {
+        prompt_with_lora = prompt;
 
         if (sample_params.sample_steps <= 0) {
             LOG_ERROR("error: the sample_steps must be greater than 0\n");
@@ -1766,7 +1758,23 @@ struct SDGenerationParams {
                     } else if (key == "Bn" || key == "bn") {
                         cache_params.Bn_compute_blocks = std::stoi(val);
                     } else if (key == "warmup") {
+                        if (cache_mode == "spectrum") {
+                            cache_params.spectrum_warmup_steps = std::stoi(val);
+                        } else {
                             cache_params.max_warmup_steps = std::stoi(val);
+                        }
+                    } else if (key == "w") {
+                        cache_params.spectrum_w = std::stof(val);
+                    } else if (key == "m") {
+                        cache_params.spectrum_m = std::stoi(val);
+                    } else if (key == "lam") {
+                        cache_params.spectrum_lam = std::stof(val);
+                    } else if (key == "window") {
+                        cache_params.spectrum_window_size = std::stoi(val);
+                    } else if (key == "flex") {
+                        cache_params.spectrum_flex_window = std::stof(val);
+                    } else if (key == "stop") {
+                        cache_params.spectrum_stop_percent = std::stof(val);
                     } else {
                         LOG_ERROR("error: unknown cache parameter '%s'", key.c_str());
                         return false;
@@ -1782,38 +1790,16 @@ struct SDGenerationParams {
         if (!cache_mode.empty()) {
             if (cache_mode == "easycache") {
                 cache_params.mode = SD_CACHE_EASYCACHE;
-                cache_params.reuse_threshold        = 0.2f;
-                cache_params.start_percent          = 0.15f;
-                cache_params.end_percent            = 0.95f;
-                cache_params.error_decay_rate       = 1.0f;
-                cache_params.use_relative_threshold = true;
-                cache_params.reset_error_on_compute = true;
             } else if (cache_mode == "ucache") {
                 cache_params.mode = SD_CACHE_UCACHE;
-                cache_params.reuse_threshold        = 1.0f;
-                cache_params.start_percent          = 0.15f;
-                cache_params.end_percent            = 0.95f;
-                cache_params.error_decay_rate       = 1.0f;
-                cache_params.use_relative_threshold = true;
-                cache_params.reset_error_on_compute = true;
             } else if (cache_mode == "dbcache") {
                 cache_params.mode = SD_CACHE_DBCACHE;
-                cache_params.Fn_compute_blocks       = 8;
-                cache_params.Bn_compute_blocks       = 0;
-                cache_params.residual_diff_threshold = 0.08f;
-                cache_params.max_warmup_steps        = 8;
             } else if (cache_mode == "taylorseer") {
                 cache_params.mode = SD_CACHE_TAYLORSEER;
-                cache_params.Fn_compute_blocks       = 8;
-                cache_params.Bn_compute_blocks       = 0;
-                cache_params.residual_diff_threshold = 0.08f;
-                cache_params.max_warmup_steps        = 8;
             } else if (cache_mode == "cache-dit") {
                 cache_params.mode = SD_CACHE_CACHE_DIT;
-                cache_params.Fn_compute_blocks       = 8;
+            } else if (cache_mode == "spectrum") {
-                cache_params.Bn_compute_blocks       = 0;
+                cache_params.mode = SD_CACHE_SPECTRUM;
-                cache_params.residual_diff_threshold = 0.08f;
-                cache_params.max_warmup_steps        = 8;
             }
 
             if (!cache_option.empty()) {
@@ -2083,6 +2069,22 @@ uint8_t* load_image_from_file(const char* image_path,
     return load_image_common(false, image_path, 0, width, height, expected_width, expected_height, expected_channel);
 }
 
+bool load_sd_image_from_file(sd_image_t* image,
+                             const char* image_path,
+                             int expected_width   = 0,
+                             int expected_height  = 0,
+                             int expected_channel = 3) {
+    int width;
+    int height;
+    image->data = load_image_common(false, image_path, 0, width, height, expected_width, expected_height, expected_channel);
+    if (image->data == nullptr) {
+        return false;
+    }
+    image->width  = width;
+    image->height = height;
+    return true;
+}
+
 uint8_t* load_image_from_memory(const char* image_bytes,
                                 int len,
                                 int& width,

examples/server/README.md

@@ -5,8 +5,8 @@ usage: ./bin/sd-server  [options]
 
 Svr Options:
   -l, --listen-ip <string>      server listen ip (default: 127.0.0.1)        
-  --listen-port <int>         server listen port (default: 1234)
   --serve-html-path <string>    path to HTML file to serve at root (optional)
+  --listen-port <int>           server listen port (default: 1234)
   -v, --verbose                 print extra info
   --color                       colors the logging tags according to level   
   -h, --help                    show this help message and exit
@@ -36,21 +36,22 @@ Context Options:
                                            CPU physical cores
   --chroma-t5-mask-pad <int>               t5 mask pad size of chroma
   --vae-tile-overlap <float>               tile overlap for vae tiling, in fraction of tile size (default: 0.5)
-  --flow-shift <float>                     shift value for Flow models like SD3.x or WAN (default: auto)
   --vae-tiling                             process vae in tiles to reduce memory usage
   --force-sdxl-vae-conv-scale              force use of conv scale on sdxl vae
   --offload-to-cpu                         place the weights in RAM to save VRAM, and automatically load them into VRAM when needed
+  --mmap                                   whether to memory-map model
   --control-net-cpu                        keep controlnet in cpu (for low vram)
   --clip-on-cpu                            keep clip in cpu (for low vram)
   --vae-on-cpu                             keep vae in cpu (for low vram)
-  --mmap                                   whether to memory-map model
+  --fa                                     use flash attention
-  --diffusion-fa                           use flash attention in the diffusion model
+  --diffusion-fa                           use flash attention in the diffusion model only
   --diffusion-conv-direct                  use ggml_conv2d_direct in the diffusion model
   --vae-conv-direct                        use ggml_conv2d_direct in the vae model
   --circular                               enable circular padding for convolutions
   --circularx                              enable circular RoPE wrapping on x-axis (width) only
   --circulary                              enable circular RoPE wrapping on y-axis (height) only
   --chroma-disable-dit-mask                disable dit mask for chroma
+  --qwen-image-zero-cond-t                 enable zero_cond_t for qwen image
   --chroma-enable-t5-mask                  enable t5 mask for chroma
   --type                                   weight type (examples: f32, f16, q4_0, q4_1, q5_0, q5_1, q8_0, q2_K, q3_K, q4_K). If not specified, the default is the
                                            type of the weight file
@@ -100,6 +101,7 @@ Default Generation Options:
   --skip-layer-start <float>               SLG enabling point (default: 0.01)
   --skip-layer-end <float>                 SLG disabling point (default: 0.2)
   --eta <float>                            eta in DDIM, only for DDIM and TCD (default: 0)
+  --flow-shift <float>                     shift value for Flow models like SD3.x or WAN (default: auto)
   --high-noise-cfg-scale <float>           (high noise) unconditional guidance scale: (default: 7.0)
   --high-noise-img-cfg-scale <float>       (high noise) image guidance scale for inpaint or instruct-pix2pix models (default: same as --cfg-scale)
   --high-noise-guidance <float>            (high noise) distilled guidance scale for models with guidance input (default: 3.5)
@@ -116,20 +118,21 @@ Default Generation Options:
   --disable-auto-resize-ref-image          disable auto resize of ref images
   -s, --seed                               RNG seed (default: 42, use random seed for < 0)
   --sampling-method                        sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing,
-                                           tcd] (default: euler for Flux/SD3/Wan, euler_a otherwise)
+                                           tcd, res_multistep, res_2s] (default: euler for Flux/SD3/Wan, euler_a
+                                           otherwise)
   --high-noise-sampling-method             (high noise) sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm,
-                                           ddim_trailing, tcd] default: euler for Flux/SD3/Wan, euler_a otherwise
+                                           ddim_trailing, tcd, res_multistep, res_2s] default: euler for Flux/SD3/Wan,
+                                           euler_a otherwise
   --scheduler                              denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits, smoothstep, sgm_uniform, simple,
-                                           kl_optimal, lcm], default: discrete
+                                           kl_optimal, lcm, bong_tangent], default: discrete
   --sigmas                                 custom sigma values for the sampler, comma-separated (e.g., "14.61,7.8,3.5,0.0").
   --skip-layers                            layers to skip for SLG steps (default: [7,8,9])
   --high-noise-skip-layers                 (high noise) layers to skip for SLG steps (default: [7,8,9])
   -r, --ref-image                          reference image for Flux Kontext models (can be used multiple times)
-  --cache-mode                             caching method: 'easycache' (DiT), 'ucache' (UNET), 'dbcache'/'taylorseer'/'cache-dit' (DiT block-level)
+  --cache-mode                             caching method: 'easycache' (DiT), 'ucache' (UNET), 'dbcache'/'taylorseer'/'cache-dit' (DiT block-level), 'spectrum' (UNET/DiT Chebyshev+Taylor forecasting)
   --cache-option                           named cache params (key=value format, comma-separated). easycache/ucache:
                                            threshold=,start=,end=,decay=,relative=,reset=; dbcache/taylorseer/cache-dit: Fn=,Bn=,threshold=,warmup=. Examples:
                                            "threshold=0.25" or "threshold=1.5,reset=0"
-  --cache-preset                           cache-dit preset: 'slow'/'s', 'medium'/'m', 'fast'/'f', 'ultra'/'u'
   --scm-mask                               SCM steps mask for cache-dit: comma-separated 0/1 (e.g., "1,1,1,0,0,1,0,0,1,0") - 1=compute, 0=can cache
   --scm-policy                             SCM policy: 'dynamic' (default) or 'static'
 ```

examples/server/main.cpp

@@ -267,6 +267,24 @@ void sd_log_cb(enum sd_log_level_t level, const char* log, void* data) {
     log_print(level, log, svr_params->verbose, svr_params->color);
 }
 
+struct LoraEntry {
+    std::string name;
+    std::string path;
+    std::string fullpath;
+};
+
+void free_results(sd_image_t* result_images, int num_results) {
+    if (result_images) {
+        for (int i = 0; i < num_results; ++i) {
+            if (result_images[i].data) {
+                stbi_image_free(result_images[i].data);
+                result_images[i].data = nullptr;
+            }
+        }
+    }
+    free(result_images);
+}
+
 int main(int argc, const char** argv) {
     if (argc > 1 && std::string(argv[1]) == "--version") {
         std::cout << version_string() << "\n";
@@ -297,6 +315,56 @@ int main(int argc, const char** argv) {
 
     std::mutex sd_ctx_mutex;
 
+    std::vector<LoraEntry> lora_cache;
+    std::mutex lora_mutex;
+
+    auto refresh_lora_cache = [&]() {
+        std::vector<LoraEntry> new_cache;
+
+        fs::path lora_dir = ctx_params.lora_model_dir;
+        if (fs::exists(lora_dir) && fs::is_directory(lora_dir)) {
+            auto is_lora_ext = [](const fs::path& p) {
+                auto ext = p.extension().string();
+                std::transform(ext.begin(), ext.end(), ext.begin(), ::tolower);
+                return ext == ".gguf" || ext == ".pt" || ext == ".pth" || ext == ".safetensors";
+            };
+
+            for (auto& entry : fs::recursive_directory_iterator(lora_dir)) {
+                if (!entry.is_regular_file())
+                    continue;
+                const fs::path& p = entry.path();
+                if (!is_lora_ext(p))
+                    continue;
+
+                LoraEntry e;
+                e.name          = p.stem().u8string();
+                e.fullpath      = p.u8string();
+                std::string rel = p.lexically_relative(lora_dir).u8string();
+                std::replace(rel.begin(), rel.end(), '\\', '/');
+                e.path = rel;
+
+                new_cache.push_back(std::move(e));
+            }
+        }
+
+        std::sort(new_cache.begin(), new_cache.end(),
+                  [](const LoraEntry& a, const LoraEntry& b) {
+                      return a.path < b.path;
+                  });
+
+        {
+            std::lock_guard<std::mutex> lock(lora_mutex);
+            lora_cache = std::move(new_cache);
+        }
+    };
+
+    auto get_lora_full_path = [&](const std::string& path) -> std::string {
+        std::lock_guard<std::mutex> lock(lora_mutex);
+        auto it = std::find_if(lora_cache.begin(), lora_cache.end(),
+                               [&](const LoraEntry& e) { return e.path == path; });
+        return (it != lora_cache.end()) ? it->fullpath : "";
+    };
+
     httplib::Server svr;
 
     svr.set_pre_routing_handler([](const httplib::Request& req, httplib::Response& res) {
@@ -361,8 +429,8 @@ int main(int argc, const char** argv) {
             std::string size          = j.value("size", "");
             std::string output_format = j.value("output_format", "png");
             int output_compression    = j.value("output_compression", 100);
-            int width                 = 512;
+            int width                 = default_gen_params.width > 0 ? default_gen_params.width : 512;
-            int height                = 512;
+            int height                = default_gen_params.width > 0 ? default_gen_params.height : 512;
             if (!size.empty()) {
                 auto pos = size.find('x');
                 if (pos != std::string::npos) {
@@ -491,6 +559,7 @@ int main(int argc, const char** argv) {
                 item["b64_json"] = b64;
                 out["data"].push_back(item);
             }
+            free_results(results, num_results);
 
             res.set_content(out.dump(), "application/json");
             res.status = 200;
@@ -522,7 +591,8 @@ int main(int argc, const char** argv) {
             std::string sd_cpp_extra_args_str = extract_and_remove_sd_cpp_extra_args(prompt);
 
             size_t image_count    = req.form.get_file_count("image[]");
-            if (image_count == 0) {
+            bool has_legacy_image = req.form.has_file("image");
+            if (image_count == 0 && !has_legacy_image) {
                 res.status = 400;
                 res.set_content(R"({"error":"at least one image[] required"})", "application/json");
                 return;
@@ -533,6 +603,10 @@ int main(int argc, const char** argv) {
                 auto file = req.form.get_file("image[]", i);
                 images_bytes.emplace_back(file.content.begin(), file.content.end());
             }
+            if (image_count == 0 && has_legacy_image) {
+                auto file = req.form.get_file("image");
+                images_bytes.emplace_back(file.content.begin(), file.content.end());
+            }
 
             std::vector<uint8_t> mask_bytes;
             if (req.form.has_file("mask")) {
@@ -550,7 +624,7 @@ int main(int argc, const char** argv) {
             n = std::clamp(n, 1, 8);
 
             std::string size = req.form.get_field("size");
-            int width = 512, height = 512;
+            int width = -1, height = -1;
             if (!size.empty()) {
                 auto pos = size.find('x');
                 if (pos != std::string::npos) {
@@ -607,15 +681,31 @@ int main(int argc, const char** argv) {
 
             LOG_DEBUG("%s\n", gen_params.to_string().c_str());
 
-            sd_image_t init_image    = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 3, nullptr};
+            sd_image_t init_image    = {0, 0, 3, nullptr};
-            sd_image_t control_image = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 3, nullptr};
+            sd_image_t control_image = {0, 0, 3, nullptr};
             std::vector<sd_image_t> pmid_images;
 
+            auto get_resolved_width = [&gen_params, &default_gen_params]() -> int {
+                if (gen_params.width > 0)
+                    return gen_params.width;
+                if (default_gen_params.width > 0)
+                    return default_gen_params.width;
+                return 512;
+            };
+            auto get_resolved_height = [&gen_params, &default_gen_params]() -> int {
+                if (gen_params.height > 0)
+                    return gen_params.height;
+                if (default_gen_params.height > 0)
+                    return default_gen_params.height;
+                return 512;
+            };
+
             std::vector<sd_image_t> ref_images;
             ref_images.reserve(images_bytes.size());
             for (auto& bytes : images_bytes) {
-                int img_w           = width;
+                int img_w;
-                int img_h           = height;
+                int img_h;
+
                 uint8_t* raw_pixels = load_image_from_memory(
                     reinterpret_cast<const char*>(bytes.data()),
                     static_cast<int>(bytes.size()),
@@ -627,22 +717,31 @@ int main(int argc, const char** argv) {
                 }
 
                 sd_image_t img{(uint32_t)img_w, (uint32_t)img_h, 3, raw_pixels};
+                gen_params.set_width_and_height_if_unset(img.width, img.height);
                 ref_images.push_back(img);
             }
 
             sd_image_t mask_image = {0};
             if (!mask_bytes.empty()) {
-                int mask_w        = width;
+                int expected_width  = 0;
-                int mask_h        = height;
+                int expected_height = 0;
+                if (gen_params.width_and_height_are_set()) {
+                    expected_width  = gen_params.width;
+                    expected_height = gen_params.height;
+                }
+                int mask_w;
+                int mask_h;
+
                 uint8_t* mask_raw = load_image_from_memory(
                     reinterpret_cast<const char*>(mask_bytes.data()),
                     static_cast<int>(mask_bytes.size()),
                     mask_w, mask_h,
-                    width, height, 1);
+                    expected_width, expected_height, 1);
                 mask_image = {(uint32_t)mask_w, (uint32_t)mask_h, 1, mask_raw};
+                gen_params.set_width_and_height_if_unset(mask_image.width, mask_image.height);
             } else {
-                mask_image.width   = width;
+                mask_image.width   = get_resolved_width();
-                mask_image.height  = height;
+                mask_image.height  = get_resolved_height();
                 mask_image.channel = 1;
                 mask_image.data    = nullptr;
             }
@@ -659,8 +758,8 @@ int main(int argc, const char** argv) {
                 gen_params.auto_resize_ref_image,
                 gen_params.increase_ref_index,
                 mask_image,
-                gen_params.width,
+                get_resolved_width(),
-                gen_params.height,
+                get_resolved_height(),
                 gen_params.sample_params,
                 gen_params.strength,
                 gen_params.seed,
@@ -705,6 +804,7 @@ int main(int argc, const char** argv) {
                 item["b64_json"] = b64;
                 out["data"].push_back(item);
             }
+            free_results(results, num_results);
 
             res.set_content(out.dump(), "application/json");
             res.status = 200;
@@ -743,8 +843,8 @@ int main(int argc, const char** argv) {
             std::string negative_prompt = j.value("negative_prompt", "");
             int width                   = j.value("width", 512);
             int height                  = j.value("height", 512);
-            int steps                   = j.value("steps", -1);
+            int steps                   = j.value("steps", default_gen_params.sample_params.sample_steps);
-            float cfg_scale             = j.value("cfg_scale", 7.f);
+            float cfg_scale             = j.value("cfg_scale", default_gen_params.sample_params.guidance.txt_cfg);
             int64_t seed                = j.value("seed", -1);
             int batch_size              = j.value("batch_size", 1);
             int clip_skip               = j.value("clip_skip", -1);
@@ -777,6 +877,38 @@ int main(int argc, const char** argv) {
                 return bad("prompt required");
             }
 
+            std::vector<sd_lora_t> sd_loras;
+            std::vector<std::string> lora_path_storage;
+
+            if (j.contains("lora") && j["lora"].is_array()) {
+                for (const auto& item : j["lora"]) {
+                    if (!item.is_object()) {
+                        continue;
+                    }
+
+                    std::string path   = item.value("path", "");
+                    float multiplier   = item.value("multiplier", 1.0f);
+                    bool is_high_noise = item.value("is_high_noise", false);
+
+                    if (path.empty()) {
+                        return bad("lora.path required");
+                    }
+
+                    std::string fullpath = get_lora_full_path(path);
+                    if (fullpath.empty()) {
+                        return bad("invalid lora path: " + path);
+                    }
+
+                    lora_path_storage.push_back(fullpath);
+                    sd_lora_t l;
+                    l.is_high_noise = is_high_noise;
+                    l.multiplier    = multiplier;
+                    l.path          = lora_path_storage.back().c_str();
+
+                    sd_loras.push_back(l);
+                }
+            }
+
             auto get_sample_method = [](std::string name) -> enum sample_method_t {
                 enum sample_method_t result = str_to_sample_method(name.c_str());
                 if (result != SAMPLE_METHOD_COUNT) return result;
@@ -795,7 +927,11 @@ int main(int argc, const char** argv) {
                     {"lcm", LCM_SAMPLE_METHOD},
                     {"ddim", DDIM_TRAILING_SAMPLE_METHOD},
                     {"dpm++ 2m", DPMPP2M_SAMPLE_METHOD},
-                    {"k_dpmpp_2m", DPMPP2M_SAMPLE_METHOD}};
+                    {"k_dpmpp_2m", DPMPP2M_SAMPLE_METHOD},
+                    {"res multistep", RES_MULTISTEP_SAMPLE_METHOD},
+                    {"k_res_multistep", RES_MULTISTEP_SAMPLE_METHOD},
+                    {"res 2s", RES_2S_SAMPLE_METHOD},
+                    {"k_res_2s", RES_2S_SAMPLE_METHOD}};
                 auto it            = hardcoded.find(name);
                 if (it != hardcoded.end()) return it->second;
                 return SAMPLE_METHOD_COUNT;
@@ -805,16 +941,13 @@ int main(int argc, const char** argv) {
 
             enum scheduler_t scheduler = str_to_scheduler(scheduler_name.c_str());
 
-            // avoid excessive resource usage
-
             SDGenerationParams gen_params             = default_gen_params;
             gen_params.prompt                         = prompt;
             gen_params.negative_prompt                = negative_prompt;
-            gen_params.width                      = width;
-            gen_params.height                     = height;
             gen_params.seed                           = seed;
             gen_params.sample_params.sample_steps     = steps;
             gen_params.batch_count                    = batch_size;
+            gen_params.sample_params.guidance.txt_cfg = cfg_scale;
 
             if (clip_skip > 0) {
                 gen_params.clip_skip = clip_skip;
@@ -828,17 +961,36 @@ int main(int argc, const char** argv) {
                 gen_params.sample_params.scheduler = scheduler;
             }
 
+            // re-read to avoid applying 512 as default before the provided
+            // images and/or server command-line
+            gen_params.width  = j.value("width", -1);
+            gen_params.height = j.value("height", -1);
+
             LOG_DEBUG("%s\n", gen_params.to_string().c_str());
 
-            sd_image_t init_image    = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 3, nullptr};
+            sd_image_t init_image    = {0, 0, 3, nullptr};
-            sd_image_t control_image = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 3, nullptr};
+            sd_image_t control_image = {0, 0, 3, nullptr};
-            sd_image_t mask_image    = {(uint32_t)gen_params.width, (uint32_t)gen_params.height, 1, nullptr};
+            sd_image_t mask_image    = {0, 0, 1, nullptr};
             std::vector<uint8_t> mask_data;
             std::vector<sd_image_t> pmid_images;
             std::vector<sd_image_t> ref_images;
 
-            if (img2img) {
+            auto get_resolved_width = [&gen_params, &default_gen_params]() -> int {
-                auto decode_image = [](sd_image_t& image, std::string encoded) -> bool {
+                if (gen_params.width > 0)
+                    return gen_params.width;
+                if (default_gen_params.width > 0)
+                    return default_gen_params.width;
+                return 512;
+            };
+            auto get_resolved_height = [&gen_params, &default_gen_params]() -> int {
+                if (gen_params.height > 0)
+                    return gen_params.height;
+                if (default_gen_params.height > 0)
+                    return default_gen_params.height;
+                return 512;
+            };
+
+            auto decode_image = [&gen_params](sd_image_t& image, std::string encoded) -> bool {
                 // remove data URI prefix if present ("data:image/png;base64,")
                 auto comma_pos = encoded.find(',');
                 if (comma_pos != std::string::npos) {
@@ -846,20 +998,29 @@ int main(int argc, const char** argv) {
                 }
                 std::vector<uint8_t> img_data = base64_decode(encoded);
                 if (!img_data.empty()) {
-                        int img_w         = image.width;
+                    int expected_width  = 0;
-                        int img_h         = image.height;
+                    int expected_height = 0;
+                    if (gen_params.width_and_height_are_set()) {
+                        expected_width  = gen_params.width;
+                        expected_height = gen_params.height;
+                    }
+                    int img_w;
+                    int img_h;
+
                     uint8_t* raw_data = load_image_from_memory(
                         (const char*)img_data.data(), (int)img_data.size(),
                         img_w, img_h,
-                            image.width, image.height, image.channel);
+                        expected_width, expected_height, image.channel);
                     if (raw_data) {
                         image = {(uint32_t)img_w, (uint32_t)img_h, image.channel, raw_data};
+                        gen_params.set_width_and_height_if_unset(image.width, image.height);
                         return true;
                     }
                 }
                 return false;
             };
 
+            if (img2img) {
                 if (j.contains("init_images") && j["init_images"].is_array() && !j["init_images"].empty()) {
                     std::string encoded = j["init_images"][0].get<std::string>();
                     decode_image(init_image, encoded);
@@ -875,13 +1036,22 @@ int main(int argc, const char** argv) {
                         }
                     }
                 } else {
-                    mask_data          = std::vector<uint8_t>(width * height, 255);
+                    int m_width        = get_resolved_width();
-                    mask_image.width   = width;
+                    int m_height       = get_resolved_height();
-                    mask_image.height  = height;
+                    mask_data          = std::vector<uint8_t>(m_width * m_height, 255);
+                    mask_image.width   = m_width;
+                    mask_image.height  = m_height;
                     mask_image.channel = 1;
                     mask_image.data    = mask_data.data();
                 }
 
+                float denoising_strength = j.value("denoising_strength", -1.f);
+                if (denoising_strength >= 0.f) {
+                    denoising_strength  = std::min(denoising_strength, 1.0f);
+                    gen_params.strength = denoising_strength;
+                }
+            }
+
             if (j.contains("extra_images") && j["extra_images"].is_array()) {
                 for (auto extra_image : j["extra_images"]) {
                     std::string encoded  = extra_image.get<std::string>();
@@ -892,16 +1062,9 @@ int main(int argc, const char** argv) {
                 }
             }
 
-                float denoising_strength = j.value("denoising_strength", -1.f);
-                if (denoising_strength >= 0.f) {
-                    denoising_strength  = std::min(denoising_strength, 1.0f);
-                    gen_params.strength = denoising_strength;
-                }
-            }
-
             sd_img_gen_params_t img_gen_params = {
-                gen_params.lora_vec.data(),
+                sd_loras.data(),
-                static_cast<uint32_t>(gen_params.lora_vec.size()),
+                static_cast<uint32_t>(sd_loras.size()),
                 gen_params.prompt.c_str(),
                 gen_params.negative_prompt.c_str(),
                 gen_params.clip_skip,
@@ -911,8 +1074,8 @@ int main(int argc, const char** argv) {
                 gen_params.auto_resize_ref_image,
                 gen_params.increase_ref_index,
                 mask_image,
-                gen_params.width,
+                get_resolved_width(),
-                gen_params.height,
+                get_resolved_height(),
                 gen_params.sample_params,
                 gen_params.strength,
                 gen_params.seed,
@@ -962,6 +1125,7 @@ int main(int argc, const char** argv) {
                 std::string b64 = base64_encode(image_bytes);
                 out["images"].push_back(b64);
             }
+            free_results(results, num_results);
 
             res.set_content(out.dump(), "application/json");
             res.status = 200;
@@ -993,6 +1157,23 @@ int main(int argc, const char** argv) {
         sdapi_any2img(req, res, true);
     });
 
+    svr.Get("/sdapi/v1/loras", [&](const httplib::Request&, httplib::Response& res) {
+        refresh_lora_cache();
+
+        json result = json::array();
+        {
+            std::lock_guard<std::mutex> lock(lora_mutex);
+            for (const auto& e : lora_cache) {
+                json item;
+                item["name"] = e.name;
+                item["path"] = e.path;
+                result.push_back(item);
+            }
+        }
+
+        res.set_content(result.dump(), "application/json");
+    });
+
     svr.Get("/sdapi/v1/samplers", [&](const httplib::Request&, httplib::Response& res) {
         std::vector<std::string> sampler_names;
         sampler_names.push_back("default");

format-code.sh

@@ -1,4 +1,4 @@
-for f in *.cpp *.h *.hpp examples/cli/*.cpp examples/common/*.hpp examples/cli/*.h examples/server/*.cpp; do
+for f in src/*.cpp src/*.h src/*.hpp src/vocab/*.h src/vocab/*.cpp examples/cli/*.cpp examples/common/*.hpp examples/cli/*.h examples/server/*.cpp; do
   [[ "$f" == vocab* ]] && continue
   echo "formatting '$f'"
   # if [ "$f" != "stable-diffusion.h" ]; then

ggml

@@ -1 +1 @@
-Subproject commit 8891ab6fc742ac1198736d3da3b73c730e42af84
+Subproject commit a8db410a252c8c8f2d120c6f2e7133ebe032f35d

include/stable-diffusion.h

@@ -48,6 +48,8 @@ enum sample_method_t {
     LCM_SAMPLE_METHOD,
     DDIM_TRAILING_SAMPLE_METHOD,
     TCD_SAMPLE_METHOD,
+    RES_MULTISTEP_SAMPLE_METHOD,
+    RES_2S_SAMPLE_METHOD,
     SAMPLE_METHOD_COUNT
 };
 
@@ -62,6 +64,7 @@ enum scheduler_t {
     SMOOTHSTEP_SCHEDULER,
     KL_OPTIMAL_SCHEDULER,
     LCM_SCHEDULER,
+    BONG_TANGENT_SCHEDULER,
     SCHEDULER_COUNT
 };
 
@@ -186,6 +189,7 @@ typedef struct {
     bool keep_clip_on_cpu;
     bool keep_control_net_on_cpu;
     bool keep_vae_on_cpu;
+    bool flash_attn;
     bool diffusion_flash_attn;
     bool tae_preview_only;
     bool diffusion_conv_direct;
@@ -197,7 +201,6 @@ typedef struct {
     bool chroma_use_t5_mask;
     int chroma_t5_mask_pad;
     bool qwen_image_zero_cond_t;
-    float flow_shift;
 } sd_ctx_params_t;
 
 typedef struct {
@@ -231,6 +234,7 @@ typedef struct {
     int shifted_timestep;
     float* custom_sigmas;
     int custom_sigmas_count;
+    float flow_shift;
 } sd_sample_params_t;
 
 typedef struct {
@@ -247,6 +251,7 @@ enum sd_cache_mode_t {
     SD_CACHE_DBCACHE,
     SD_CACHE_TAYLORSEER,
     SD_CACHE_CACHE_DIT,
+    SD_CACHE_SPECTRUM,
 };
 
 typedef struct {
@@ -267,6 +272,13 @@ typedef struct {
     int taylorseer_skip_interval;
     const char* scm_mask;
     bool scm_policy_dynamic;
+    float spectrum_w;
+    int spectrum_m;
+    float spectrum_lam;
+    int spectrum_window_size;
+    float spectrum_flex_window;
+    int spectrum_warmup_steps;
+    float spectrum_stop_percent;
 } sd_cache_params_t;
 
 typedef struct {

src/anima.hpp

@@ -0,0 +1,686 @@
+#ifndef __ANIMA_HPP__
+#define __ANIMA_HPP__
+
+#include <cmath>
+#include <memory>
+#include <utility>
+#include <vector>
+
+#include "common_block.hpp"
+#include "flux.hpp"
+#include "rope.hpp"
+
+namespace Anima {
+    constexpr int ANIMA_GRAPH_SIZE = 65536;
+
+    __STATIC_INLINE__ struct ggml_tensor* apply_gate(struct ggml_context* ctx,
+                                                     struct ggml_tensor* x,
+                                                     struct ggml_tensor* gate) {
+        gate = ggml_reshape_3d(ctx, gate, gate->ne[0], 1, gate->ne[1]);  // [N, 1, C]
+        return ggml_mul(ctx, x, gate);
+    }
+
+    struct XEmbedder : public GGMLBlock {
+    public:
+        XEmbedder(int64_t in_dim, int64_t out_dim) {
+            blocks["proj.1"] = std::make_shared<Linear>(in_dim, out_dim, false);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x) {
+            auto proj = std::dynamic_pointer_cast<Linear>(blocks["proj.1"]);
+            return proj->forward(ctx, x);
+        }
+    };
+
+    struct TimestepEmbedder : public GGMLBlock {
+    public:
+        TimestepEmbedder(int64_t in_dim, int64_t out_dim) {
+            blocks["1.linear_1"] = std::make_shared<Linear>(in_dim, in_dim, false);
+            blocks["1.linear_2"] = std::make_shared<Linear>(in_dim, out_dim, false);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x) {
+            auto linear_1 = std::dynamic_pointer_cast<Linear>(blocks["1.linear_1"]);
+            auto linear_2 = std::dynamic_pointer_cast<Linear>(blocks["1.linear_2"]);
+
+            x = linear_1->forward(ctx, x);
+            x = ggml_silu_inplace(ctx->ggml_ctx, x);
+            x = linear_2->forward(ctx, x);
+            return x;
+        }
+    };
+
+    struct AdaLayerNormZero : public GGMLBlock {
+    protected:
+        int64_t in_features;
+
+    public:
+        AdaLayerNormZero(int64_t in_features, int64_t hidden_features = 256)
+            : in_features(in_features) {
+            blocks["norm"] = std::make_shared<LayerNorm>(in_features, 1e-6f, false, false);
+            blocks["1"]    = std::make_shared<Linear>(in_features, hidden_features, false);
+            blocks["2"]    = std::make_shared<Linear>(hidden_features, 3 * in_features, false);
+        }
+
+        std::pair<struct ggml_tensor*, struct ggml_tensor*> forward(GGMLRunnerContext* ctx,
+                                                                    struct ggml_tensor* hidden_states,
+                                                                    struct ggml_tensor* embedded_timestep,
+                                                                    struct ggml_tensor* temb = nullptr) {
+            auto norm     = std::dynamic_pointer_cast<LayerNorm>(blocks["norm"]);
+            auto linear_1 = std::dynamic_pointer_cast<Linear>(blocks["1"]);
+            auto linear_2 = std::dynamic_pointer_cast<Linear>(blocks["2"]);
+
+            auto emb = ggml_silu(ctx->ggml_ctx, embedded_timestep);
+            emb      = linear_1->forward(ctx, emb);
+            emb      = linear_2->forward(ctx, emb);  // [N, 3*C]
+
+            if (temb != nullptr) {
+                emb = ggml_add(ctx->ggml_ctx, emb, temb);
+            }
+
+            auto emb_chunks = ggml_ext_chunk(ctx->ggml_ctx, emb, 3, 0);
+            auto shift      = emb_chunks[0];
+            auto scale      = emb_chunks[1];
+            auto gate       = emb_chunks[2];
+
+            auto x = norm->forward(ctx, hidden_states);
+            x      = Flux::modulate(ctx->ggml_ctx, x, shift, scale);
+
+            return {x, gate};
+        }
+    };
+
+    struct AdaLayerNorm : public GGMLBlock {
+    protected:
+        int64_t embedding_dim;
+
+    public:
+        AdaLayerNorm(int64_t in_features, int64_t hidden_features = 256)
+            : embedding_dim(in_features) {
+            blocks["norm"] = std::make_shared<LayerNorm>(in_features, 1e-6f, false, false);
+            blocks["1"]    = std::make_shared<Linear>(in_features, hidden_features, false);
+            blocks["2"]    = std::make_shared<Linear>(hidden_features, 2 * in_features, false);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx,
+                                    struct ggml_tensor* hidden_states,
+                                    struct ggml_tensor* embedded_timestep,
+                                    struct ggml_tensor* temb = nullptr) {
+            auto norm     = std::dynamic_pointer_cast<LayerNorm>(blocks["norm"]);
+            auto linear_1 = std::dynamic_pointer_cast<Linear>(blocks["1"]);
+            auto linear_2 = std::dynamic_pointer_cast<Linear>(blocks["2"]);
+
+            auto emb = ggml_silu(ctx->ggml_ctx, embedded_timestep);
+            emb      = linear_1->forward(ctx, emb);
+            emb      = linear_2->forward(ctx, emb);  // [N, 2*C]
+
+            if (temb != nullptr) {
+                auto temb_2c = ggml_view_2d(ctx->ggml_ctx, temb, 2 * embedding_dim, temb->ne[1], temb->nb[1], 0);
+                emb          = ggml_add(ctx->ggml_ctx, emb, temb_2c);
+            }
+
+            auto emb_chunks = ggml_ext_chunk(ctx->ggml_ctx, emb, 2, 0);
+            auto shift      = emb_chunks[0];
+            auto scale      = emb_chunks[1];
+
+            auto x = norm->forward(ctx, hidden_states);
+            x      = Flux::modulate(ctx->ggml_ctx, x, shift, scale);
+            return x;
+        }
+    };
+
+    struct AnimaAttention : public GGMLBlock {
+    protected:
+        int64_t num_heads;
+        int64_t head_dim;
+        std::string out_proj_name;
+
+    public:
+        AnimaAttention(int64_t query_dim,
+                       int64_t context_dim,
+                       int64_t num_heads,
+                       int64_t head_dim,
+                       const std::string& out_proj_name = "output_proj")
+            : num_heads(num_heads), head_dim(head_dim), out_proj_name(out_proj_name) {
+            int64_t inner_dim = num_heads * head_dim;
+
+            blocks["q_proj"]            = std::make_shared<Linear>(query_dim, inner_dim, false);
+            blocks["k_proj"]            = std::make_shared<Linear>(context_dim, inner_dim, false);
+            blocks["v_proj"]            = std::make_shared<Linear>(context_dim, inner_dim, false);
+            blocks["q_norm"]            = std::make_shared<RMSNorm>(head_dim, 1e-6f);
+            blocks["k_norm"]            = std::make_shared<RMSNorm>(head_dim, 1e-6f);
+            blocks[this->out_proj_name] = std::make_shared<Linear>(inner_dim, query_dim, false);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx,
+                                    struct ggml_tensor* hidden_states,
+                                    struct ggml_tensor* encoder_hidden_states = nullptr,
+                                    struct ggml_tensor* pe_q                  = nullptr,
+                                    struct ggml_tensor* pe_k                  = nullptr) {
+            if (encoder_hidden_states == nullptr) {
+                encoder_hidden_states = hidden_states;
+            }
+
+            auto q_proj   = std::dynamic_pointer_cast<Linear>(blocks["q_proj"]);
+            auto k_proj   = std::dynamic_pointer_cast<Linear>(blocks["k_proj"]);
+            auto v_proj   = std::dynamic_pointer_cast<Linear>(blocks["v_proj"]);
+            auto q_norm   = std::dynamic_pointer_cast<RMSNorm>(blocks["q_norm"]);
+            auto k_norm   = std::dynamic_pointer_cast<RMSNorm>(blocks["k_norm"]);
+            auto out_proj = std::dynamic_pointer_cast<Linear>(blocks[out_proj_name]);
+
+            auto q = q_proj->forward(ctx, hidden_states);
+            auto k = k_proj->forward(ctx, encoder_hidden_states);
+            auto v = v_proj->forward(ctx, encoder_hidden_states);
+
+            int64_t N   = q->ne[2];
+            int64_t L_q = q->ne[1];
+            int64_t L_k = k->ne[1];
+
+            auto q4 = ggml_reshape_4d(ctx->ggml_ctx, q, head_dim, num_heads, L_q, N);  // [N, L_q, H, D]
+            auto k4 = ggml_reshape_4d(ctx->ggml_ctx, k, head_dim, num_heads, L_k, N);  // [N, L_k, H, D]
+            auto v4 = ggml_reshape_4d(ctx->ggml_ctx, v, head_dim, num_heads, L_k, N);  // [N, L_k, H, D]
+
+            q4 = q_norm->forward(ctx, q4);
+            k4 = k_norm->forward(ctx, k4);
+
+            struct ggml_tensor* attn_out = nullptr;
+            if (pe_q != nullptr || pe_k != nullptr) {
+                if (pe_q == nullptr) {
+                    pe_q = pe_k;
+                }
+                if (pe_k == nullptr) {
+                    pe_k = pe_q;
+                }
+                auto q_rope = Rope::apply_rope(ctx->ggml_ctx, q4, pe_q, false);
+                auto k_rope = Rope::apply_rope(ctx->ggml_ctx, k4, pe_k, false);
+                attn_out    = ggml_ext_attention_ext(ctx->ggml_ctx,
+                                                     ctx->backend,
+                                                     q_rope,
+                                                     k_rope,
+                                                     v4,
+                                                     num_heads,
+                                                     nullptr,
+                                                     true,
+                                                     ctx->flash_attn_enabled);
+            } else {
+                auto q_flat = ggml_reshape_3d(ctx->ggml_ctx, q4, head_dim * num_heads, L_q, N);
+                auto k_flat = ggml_reshape_3d(ctx->ggml_ctx, k4, head_dim * num_heads, L_k, N);
+                attn_out    = ggml_ext_attention_ext(ctx->ggml_ctx,
+                                                     ctx->backend,
+                                                     q_flat,
+                                                     k_flat,
+                                                     v,
+                                                     num_heads,
+                                                     nullptr,
+                                                     false,
+                                                     ctx->flash_attn_enabled);
+            }
+
+            return out_proj->forward(ctx, attn_out);
+        }
+    };
+
+    struct AnimaMLP : public GGMLBlock {
+    public:
+        AnimaMLP(int64_t dim, int64_t hidden_dim) {
+            blocks["layer1"] = std::make_shared<Linear>(dim, hidden_dim, false);
+            blocks["layer2"] = std::make_shared<Linear>(hidden_dim, dim, false);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x) {
+            auto layer1 = std::dynamic_pointer_cast<Linear>(blocks["layer1"]);
+            auto layer2 = std::dynamic_pointer_cast<Linear>(blocks["layer2"]);
+
+            x = layer1->forward(ctx, x);
+            x = ggml_ext_gelu(ctx->ggml_ctx, x, true);
+            x = layer2->forward(ctx, x);
+            return x;
+        }
+    };
+
+    struct AdapterMLP : public GGMLBlock {
+    public:
+        AdapterMLP(int64_t dim, int64_t hidden_dim) {
+            blocks["0"] = std::make_shared<Linear>(dim, hidden_dim, true);
+            blocks["2"] = std::make_shared<Linear>(hidden_dim, dim, true);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x) {
+            auto layer0 = std::dynamic_pointer_cast<Linear>(blocks["0"]);
+            auto layer2 = std::dynamic_pointer_cast<Linear>(blocks["2"]);
+
+            x = layer0->forward(ctx, x);
+            x = ggml_ext_gelu(ctx->ggml_ctx, x, true);
+            x = layer2->forward(ctx, x);
+            return x;
+        }
+    };
+
+    struct LLMAdapterBlock : public GGMLBlock {
+    public:
+        LLMAdapterBlock(int64_t model_dim = 1024, int64_t source_dim = 1024, int64_t num_heads = 16, int64_t head_dim = 64) {
+            blocks["norm_self_attn"]  = std::make_shared<RMSNorm>(model_dim, 1e-6f);
+            blocks["self_attn"]       = std::make_shared<AnimaAttention>(model_dim, model_dim, num_heads, head_dim, "o_proj");
+            blocks["norm_cross_attn"] = std::make_shared<RMSNorm>(model_dim, 1e-6f);
+            blocks["cross_attn"]      = std::make_shared<AnimaAttention>(model_dim, source_dim, num_heads, head_dim, "o_proj");
+            blocks["norm_mlp"]        = std::make_shared<RMSNorm>(model_dim, 1e-6f);
+            blocks["mlp"]             = std::make_shared<AdapterMLP>(model_dim, model_dim * 4);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx,
+                                    struct ggml_tensor* x,
+                                    struct ggml_tensor* context,
+                                    struct ggml_tensor* target_pe,
+                                    struct ggml_tensor* context_pe) {
+            auto norm_self_attn  = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_self_attn"]);
+            auto self_attn       = std::dynamic_pointer_cast<AnimaAttention>(blocks["self_attn"]);
+            auto norm_cross_attn = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_cross_attn"]);
+            auto cross_attn      = std::dynamic_pointer_cast<AnimaAttention>(blocks["cross_attn"]);
+            auto norm_mlp        = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_mlp"]);
+            auto mlp             = std::dynamic_pointer_cast<AdapterMLP>(blocks["mlp"]);
+
+            auto h = norm_self_attn->forward(ctx, x);
+            h      = self_attn->forward(ctx, h, nullptr, target_pe, target_pe);
+            x      = ggml_add(ctx->ggml_ctx, x, h);
+
+            h = norm_cross_attn->forward(ctx, x);
+            h = cross_attn->forward(ctx, h, context, target_pe, context_pe);
+            x = ggml_add(ctx->ggml_ctx, x, h);
+
+            h = norm_mlp->forward(ctx, x);
+            h = mlp->forward(ctx, h);
+            x = ggml_add(ctx->ggml_ctx, x, h);
+
+            return x;
+        }
+    };
+
+    struct LLMAdapter : public GGMLBlock {
+    protected:
+        int num_layers;
+
+    public:
+        LLMAdapter(int64_t source_dim = 1024,
+                   int64_t target_dim = 1024,
+                   int64_t model_dim  = 1024,
+                   int num_layers     = 6,
+                   int num_heads      = 16)
+            : num_layers(num_layers) {
+            int64_t head_dim = model_dim / num_heads;
+
+            blocks["embed"] = std::make_shared<Embedding>(32128, target_dim);
+            for (int i = 0; i < num_layers; i++) {
+                blocks["blocks." + std::to_string(i)] =
+                    std::make_shared<LLMAdapterBlock>(model_dim, source_dim, num_heads, head_dim);
+            }
+            blocks["out_proj"] = std::make_shared<Linear>(model_dim, target_dim, true);
+            blocks["norm"]     = std::make_shared<RMSNorm>(target_dim, 1e-6f);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx,
+                                    struct ggml_tensor* source_hidden_states,
+                                    struct ggml_tensor* target_input_ids,
+                                    struct ggml_tensor* target_pe,
+                                    struct ggml_tensor* source_pe) {
+            GGML_ASSERT(target_input_ids != nullptr);
+            if (ggml_n_dims(target_input_ids) == 1) {
+                target_input_ids = ggml_reshape_2d(ctx->ggml_ctx, target_input_ids, target_input_ids->ne[0], 1);
+            }
+
+            auto embed    = std::dynamic_pointer_cast<Embedding>(blocks["embed"]);
+            auto out_proj = std::dynamic_pointer_cast<Linear>(blocks["out_proj"]);
+            auto norm     = std::dynamic_pointer_cast<RMSNorm>(blocks["norm"]);
+
+            auto x = embed->forward(ctx, target_input_ids);  // [N, target_len, target_dim]
+
+            for (int i = 0; i < num_layers; i++) {
+                auto block = std::dynamic_pointer_cast<LLMAdapterBlock>(blocks["blocks." + std::to_string(i)]);
+                x          = block->forward(ctx, x, source_hidden_states, target_pe, source_pe);
+            }
+
+            x = out_proj->forward(ctx, x);
+            x = norm->forward(ctx, x);
+            return x;
+        }
+    };
+
+    struct TransformerBlock : public GGMLBlock {
+    public:
+        TransformerBlock(int64_t hidden_size,
+                         int64_t text_embed_dim,
+                         int64_t num_heads,
+                         int64_t head_dim,
+                         int64_t mlp_ratio      = 4,
+                         int64_t adaln_lora_dim = 256) {
+            blocks["adaln_modulation_self_attn"]  = std::make_shared<AdaLayerNormZero>(hidden_size, adaln_lora_dim);
+            blocks["self_attn"]                   = std::make_shared<AnimaAttention>(hidden_size, hidden_size, num_heads, head_dim);
+            blocks["adaln_modulation_cross_attn"] = std::make_shared<AdaLayerNormZero>(hidden_size, adaln_lora_dim);
+            blocks["cross_attn"]                  = std::make_shared<AnimaAttention>(hidden_size, text_embed_dim, num_heads, head_dim);
+            blocks["adaln_modulation_mlp"]        = std::make_shared<AdaLayerNormZero>(hidden_size, adaln_lora_dim);
+            blocks["mlp"]                         = std::make_shared<AnimaMLP>(hidden_size, hidden_size * mlp_ratio);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx,
+                                    struct ggml_tensor* hidden_states,
+                                    struct ggml_tensor* encoder_hidden_states,
+                                    struct ggml_tensor* embedded_timestep,
+                                    struct ggml_tensor* temb,
+                                    struct ggml_tensor* image_pe) {
+            auto norm1 = std::dynamic_pointer_cast<AdaLayerNormZero>(blocks["adaln_modulation_self_attn"]);
+            auto attn1 = std::dynamic_pointer_cast<AnimaAttention>(blocks["self_attn"]);
+            auto norm2 = std::dynamic_pointer_cast<AdaLayerNormZero>(blocks["adaln_modulation_cross_attn"]);
+            auto attn2 = std::dynamic_pointer_cast<AnimaAttention>(blocks["cross_attn"]);
+            auto norm3 = std::dynamic_pointer_cast<AdaLayerNormZero>(blocks["adaln_modulation_mlp"]);
+            auto mlp   = std::dynamic_pointer_cast<AnimaMLP>(blocks["mlp"]);
+
+            auto [normed1, gate1] = norm1->forward(ctx, hidden_states, embedded_timestep, temb);
+            auto h                = attn1->forward(ctx, normed1, nullptr, image_pe, image_pe);
+            hidden_states         = ggml_add(ctx->ggml_ctx, hidden_states, apply_gate(ctx->ggml_ctx, h, gate1));
+
+            auto [normed2, gate2] = norm2->forward(ctx, hidden_states, embedded_timestep, temb);
+            h                     = attn2->forward(ctx, normed2, encoder_hidden_states, nullptr, nullptr);
+            hidden_states         = ggml_add(ctx->ggml_ctx, hidden_states, apply_gate(ctx->ggml_ctx, h, gate2));
+
+            auto [normed3, gate3] = norm3->forward(ctx, hidden_states, embedded_timestep, temb);
+            h                     = mlp->forward(ctx, normed3);
+            hidden_states         = ggml_add(ctx->ggml_ctx, hidden_states, apply_gate(ctx->ggml_ctx, h, gate3));
+
+            return hidden_states;
+        }
+    };
+
+    struct FinalLayer : public GGMLBlock {
+    protected:
+        int64_t hidden_size;
+        int64_t patch_size;
+        int64_t out_channels;
+
+    public:
+        FinalLayer(int64_t hidden_size, int64_t patch_size, int64_t out_channels)
+            : hidden_size(hidden_size), patch_size(patch_size), out_channels(out_channels) {
+            blocks["adaln_modulation"] = std::make_shared<AdaLayerNorm>(hidden_size, 256);
+            blocks["linear"]           = std::make_shared<Linear>(hidden_size, patch_size * patch_size * out_channels, false);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx,
+                                    struct ggml_tensor* hidden_states,
+                                    struct ggml_tensor* embedded_timestep,
+                                    struct ggml_tensor* temb) {
+            auto adaln  = std::dynamic_pointer_cast<AdaLayerNorm>(blocks["adaln_modulation"]);
+            auto linear = std::dynamic_pointer_cast<Linear>(blocks["linear"]);
+
+            hidden_states = adaln->forward(ctx, hidden_states, embedded_timestep, temb);
+            hidden_states = linear->forward(ctx, hidden_states);
+            return hidden_states;
+        }
+    };
+
+    struct AnimaNet : public GGMLBlock {
+    public:
+        int64_t in_channels       = 16;
+        int64_t out_channels      = 16;
+        int64_t hidden_size       = 2048;
+        int64_t text_embed_dim    = 1024;
+        int64_t num_heads         = 16;
+        int64_t head_dim          = 128;
+        int patch_size            = 2;
+        int64_t num_layers        = 28;
+        std::vector<int> axes_dim = {44, 42, 42};
+        int theta                 = 10000;
+
+    public:
+        AnimaNet() = default;
+        explicit AnimaNet(int64_t num_layers)
+            : num_layers(num_layers) {
+            blocks["x_embedder"]       = std::make_shared<XEmbedder>((in_channels + 1) * patch_size * patch_size, hidden_size);
+            blocks["t_embedder"]       = std::make_shared<TimestepEmbedder>(hidden_size, hidden_size * 3);
+            blocks["t_embedding_norm"] = std::make_shared<RMSNorm>(hidden_size, 1e-6f);
+            for (int i = 0; i < num_layers; i++) {
+                blocks["blocks." + std::to_string(i)] = std::make_shared<TransformerBlock>(hidden_size,
+                                                                                           text_embed_dim,
+                                                                                           num_heads,
+                                                                                           head_dim);
+            }
+            blocks["final_layer"] = std::make_shared<FinalLayer>(hidden_size, patch_size, out_channels);
+            blocks["llm_adapter"] = std::make_shared<LLMAdapter>(1024, 1024, 1024, 6, 16);
+        }
+
+        struct ggml_tensor* forward(GGMLRunnerContext* ctx,
+                                    struct ggml_tensor* x,
+                                    struct ggml_tensor* timestep,
+                                    struct ggml_tensor* encoder_hidden_states,
+                                    struct ggml_tensor* image_pe,
+                                    struct ggml_tensor* t5_ids       = nullptr,
+                                    struct ggml_tensor* t5_weights   = nullptr,
+                                    struct ggml_tensor* adapter_q_pe = nullptr,
+                                    struct ggml_tensor* adapter_k_pe = nullptr) {
+            GGML_ASSERT(x->ne[3] == 1);
+
+            auto x_embedder       = std::dynamic_pointer_cast<XEmbedder>(blocks["x_embedder"]);
+            auto t_embedder       = std::dynamic_pointer_cast<TimestepEmbedder>(blocks["t_embedder"]);
+            auto t_embedding_norm = std::dynamic_pointer_cast<RMSNorm>(blocks["t_embedding_norm"]);
+            auto final_layer      = std::dynamic_pointer_cast<FinalLayer>(blocks["final_layer"]);
+            auto llm_adapter      = std::dynamic_pointer_cast<LLMAdapter>(blocks["llm_adapter"]);
+
+            int64_t W = x->ne[0];
+            int64_t H = x->ne[1];
+
+            auto padding_mask = ggml_ext_zeros(ctx->ggml_ctx, x->ne[0], x->ne[1], 1, x->ne[3]);
+            x                 = ggml_concat(ctx->ggml_ctx, x, padding_mask, 2);  // [N, C + 1, H, W]
+
+            x = DiT::pad_and_patchify(ctx, x, patch_size, patch_size);  // [N, h*w, (C+1)*ph*pw]
+
+            x = x_embedder->forward(ctx, x);
+
+            auto timestep_proj     = ggml_ext_timestep_embedding(ctx->ggml_ctx, timestep, static_cast<int>(hidden_size));
+            auto temb              = t_embedder->forward(ctx, timestep_proj);
+            auto embedded_timestep = t_embedding_norm->forward(ctx, timestep_proj);
+
+            if (t5_ids != nullptr) {
+                auto adapted_context = llm_adapter->forward(ctx, encoder_hidden_states, t5_ids, adapter_q_pe, adapter_k_pe);
+                if (t5_weights != nullptr) {
+                    auto w = t5_weights;
+                    if (ggml_n_dims(w) == 1) {
+                        w = ggml_reshape_3d(ctx->ggml_ctx, w, 1, w->ne[0], 1);
+                    }
+                    w               = ggml_repeat_4d(ctx->ggml_ctx, w, adapted_context->ne[0], adapted_context->ne[1], adapted_context->ne[2], 1);
+                    adapted_context = ggml_mul(ctx->ggml_ctx, adapted_context, w);
+                }
+                if (adapted_context->ne[1] < 512) {
+                    auto pad_ctx    = ggml_ext_zeros(ctx->ggml_ctx,
+                                                     adapted_context->ne[0],
+                                                     512 - adapted_context->ne[1],
+                                                     adapted_context->ne[2],
+                                                     1);
+                    adapted_context = ggml_concat(ctx->ggml_ctx, adapted_context, pad_ctx, 1);
+                } else if (adapted_context->ne[1] > 512) {
+                    adapted_context = ggml_ext_slice(ctx->ggml_ctx, adapted_context, 1, 0, 512);
+                }
+                encoder_hidden_states = adapted_context;
+            }
+
+            for (int i = 0; i < num_layers; i++) {
+                auto block = std::dynamic_pointer_cast<TransformerBlock>(blocks["blocks." + std::to_string(i)]);
+                x          = block->forward(ctx, x, encoder_hidden_states, embedded_timestep, temb, image_pe);
+            }
+
+            x = final_layer->forward(ctx, x, embedded_timestep, temb);  // [N, h*w, ph*pw*C]
+
+            x = DiT::unpatchify_and_crop(ctx->ggml_ctx, x, H, W, patch_size, patch_size, false);  // [N, C, H, W]
+
+            return x;
+        }
+    };
+
+    struct AnimaRunner : public GGMLRunner {
+    public:
+        std::vector<float> image_pe_vec;
+        std::vector<float> adapter_q_pe_vec;
+        std::vector<float> adapter_k_pe_vec;
+        AnimaNet net;
+
+        AnimaRunner(ggml_backend_t backend,
+                    bool offload_params_to_cpu,
+                    const String2TensorStorage& tensor_storage_map = {},
+                    const std::string prefix                       = "model.diffusion_model")
+            : GGMLRunner(backend, offload_params_to_cpu) {
+            int64_t num_layers    = 0;
+            std::string layer_tag = prefix + ".net.blocks.";
+            for (const auto& kv : tensor_storage_map) {
+                const std::string& tensor_name = kv.first;
+                size_t pos                     = tensor_name.find(layer_tag);
+                if (pos == std::string::npos) {
+                    continue;
+                }
+                size_t start = pos + layer_tag.size();
+                size_t end   = tensor_name.find('.', start);
+                if (end == std::string::npos) {
+                    continue;
+                }
+                int64_t layer_id = atoll(tensor_name.substr(start, end - start).c_str());
+                num_layers       = std::max(num_layers, layer_id + 1);
+            }
+            if (num_layers <= 0) {
+                num_layers = 28;
+            }
+            LOG_INFO("anima net layers: %" PRId64, num_layers);
+
+            net = AnimaNet(num_layers);
+            net.init(params_ctx, tensor_storage_map, prefix + ".net");
+        }
+
+        std::string get_desc() override {
+            return "anima";
+        }
+
+        void get_param_tensors(std::map<std::string, struct ggml_tensor*>& tensors, const std::string prefix) {
+            net.get_param_tensors(tensors, prefix + ".net");
+        }
+
+        static std::vector<float> gen_1d_rope_pe_vec(int64_t seq_len, int dim, float theta = 10000.f) {
+            std::vector<float> pos(seq_len);
+            for (int64_t i = 0; i < seq_len; i++) {
+                pos[i] = static_cast<float>(i);
+            }
+            auto rope_emb = Rope::rope(pos, dim, theta);
+            return Rope::flatten(rope_emb);
+        }
+
+        static float calc_ntk_factor(float extrapolation_ratio, int axis_dim) {
+            if (extrapolation_ratio == 1.0f || axis_dim <= 2) {
+                return 1.0f;
+            }
+            return std::pow(extrapolation_ratio, static_cast<float>(axis_dim) / static_cast<float>(axis_dim - 2));
+        }
+
+        static std::vector<float> gen_anima_image_pe_vec(int bs,
+                                                         int h,
+                                                         int w,
+                                                         int patch_size,
+                                                         int theta,
+                                                         const std::vector<int>& axes_dim,
+                                                         float h_extrapolation_ratio,
+                                                         float w_extrapolation_ratio,
+                                                         float t_extrapolation_ratio) {
+            static const std::vector<ggml_tensor*> empty_ref_latents;
+            auto ids = Rope::gen_flux_ids(h,
+                                          w,
+                                          patch_size,
+                                          bs,
+                                          static_cast<int>(axes_dim.size()),
+                                          0,
+                                          {},
+                                          empty_ref_latents,
+                                          false,
+                                          1.0f);
+
+            std::vector<float> axis_thetas = {
+                static_cast<float>(theta) * calc_ntk_factor(t_extrapolation_ratio, axes_dim[0]),
+                static_cast<float>(theta) * calc_ntk_factor(h_extrapolation_ratio, axes_dim[1]),
+                static_cast<float>(theta) * calc_ntk_factor(w_extrapolation_ratio, axes_dim[2]),
+            };
+            return Rope::embed_nd(ids, bs, axis_thetas, axes_dim);
+        }
+
+        struct ggml_cgraph* build_graph(struct ggml_tensor* x,
+                                        struct ggml_tensor* timesteps,
+                                        struct ggml_tensor* context,
+                                        struct ggml_tensor* t5_ids     = nullptr,
+                                        struct ggml_tensor* t5_weights = nullptr) {
+            GGML_ASSERT(x->ne[3] == 1);
+            struct ggml_cgraph* gf = new_graph_custom(ANIMA_GRAPH_SIZE);
+
+            x          = to_backend(x);
+            timesteps  = to_backend(timesteps);
+            context    = to_backend(context);
+            t5_ids     = to_backend(t5_ids);
+            t5_weights = to_backend(t5_weights);
+
+            int64_t pad_h = (net.patch_size - x->ne[1] % net.patch_size) % net.patch_size;
+            int64_t pad_w = (net.patch_size - x->ne[0] % net.patch_size) % net.patch_size;
+            int64_t h_pad = x->ne[1] + pad_h;
+            int64_t w_pad = x->ne[0] + pad_w;
+
+            image_pe_vec          = gen_anima_image_pe_vec(1,
+                                                           static_cast<int>(h_pad),
+                                                           static_cast<int>(w_pad),
+                                                           static_cast<int>(net.patch_size),
+                                                           net.theta,
+                                                           net.axes_dim,
+                                                           4.0f,
+                                                           4.0f,
+                                                           1.0f);
+            int64_t image_pos_len = static_cast<int64_t>(image_pe_vec.size()) / (2 * 2 * (net.head_dim / 2));
+            auto image_pe         = ggml_new_tensor_4d(compute_ctx, GGML_TYPE_F32, 2, 2, net.head_dim / 2, image_pos_len);
+            set_backend_tensor_data(image_pe, image_pe_vec.data());
+
+            ggml_tensor* adapter_q_pe = nullptr;
+            ggml_tensor* adapter_k_pe = nullptr;
+            if (t5_ids != nullptr) {
+                int64_t target_len = t5_ids->ne[0];
+                int64_t source_len = context->ne[1];
+
+                adapter_q_pe_vec = gen_1d_rope_pe_vec(target_len, 64, 10000.f);
+                adapter_k_pe_vec = gen_1d_rope_pe_vec(source_len, 64, 10000.f);
+
+                int64_t target_pos_len = static_cast<int64_t>(adapter_q_pe_vec.size()) / (2 * 2 * 32);
+                int64_t source_pos_len = static_cast<int64_t>(adapter_k_pe_vec.size()) / (2 * 2 * 32);
+
+                adapter_q_pe = ggml_new_tensor_4d(compute_ctx, GGML_TYPE_F32, 2, 2, 32, target_pos_len);
+                adapter_k_pe = ggml_new_tensor_4d(compute_ctx, GGML_TYPE_F32, 2, 2, 32, source_pos_len);
+                set_backend_tensor_data(adapter_q_pe, adapter_q_pe_vec.data());
+                set_backend_tensor_data(adapter_k_pe, adapter_k_pe_vec.data());
+            }
+
+            auto runner_ctx = get_context();
+            auto out        = net.forward(&runner_ctx,
+                                          x,
+                                          timesteps,
+                                          context,
+                                          image_pe,
+                                          t5_ids,
+                                          t5_weights,
+                                          adapter_q_pe,
+                                          adapter_k_pe);
+
+            ggml_build_forward_expand(gf, out);
+            return gf;
+        }
+
+        bool compute(int n_threads,
+                     struct ggml_tensor* x,
+                     struct ggml_tensor* timesteps,
+                     struct ggml_tensor* context,
+                     struct ggml_tensor* t5_ids      = nullptr,
+                     struct ggml_tensor* t5_weights  = nullptr,
+                     struct ggml_tensor** output     = nullptr,
+                     struct ggml_context* output_ctx = nullptr) {
+            auto get_graph = [&]() -> struct ggml_cgraph* {
+                return build_graph(x, timesteps, context, t5_ids, t5_weights);
+            };
+            return GGMLRunner::compute(get_graph, n_threads, false, output, output_ctx);
+        }
+    };
+}  // namespace Anima
+
+#endif  // __ANIMA_HPP__

src/cache_dit.hpp

@@ -603,87 +603,6 @@ inline std::vector<int> generate_scm_mask(
     return mask;
 }
 
-inline std::vector<int> get_scm_preset(const std::string& preset, int total_steps) {
-    struct Preset {
-        std::vector<int> compute_bins;
-        std::vector<int> cache_bins;
-    };
-
-    Preset slow   = {{8, 3, 3, 2, 1, 1}, {1, 2, 2, 2, 3}};
-    Preset medium = {{6, 2, 2, 2, 2, 1}, {1, 3, 3, 3, 3}};
-    Preset fast   = {{6, 1, 1, 1, 1, 1}, {1, 3, 4, 5, 4}};
-    Preset ultra  = {{4, 1, 1, 1, 1}, {2, 5, 6, 7}};
-
-    Preset* p = nullptr;
-    if (preset == "slow" || preset == "s" || preset == "S")
-        p = &slow;
-    else if (preset == "medium" || preset == "m" || preset == "M")
-        p = &medium;
-    else if (preset == "fast" || preset == "f" || preset == "F")
-        p = &fast;
-    else if (preset == "ultra" || preset == "u" || preset == "U")
-        p = &ultra;
-    else
-        return {};
-
-    if (total_steps != 28 && total_steps > 0) {
-        float scale = static_cast<float>(total_steps) / 28.0f;
-        std::vector<int> scaled_compute, scaled_cache;
-
-        for (int v : p->compute_bins) {
-            scaled_compute.push_back(std::max(1, static_cast<int>(v * scale + 0.5f)));
-        }
-        for (int v : p->cache_bins) {
-            scaled_cache.push_back(std::max(1, static_cast<int>(v * scale + 0.5f)));
-        }
-
-        return generate_scm_mask(scaled_compute, scaled_cache, total_steps);
-    }
-
-    return generate_scm_mask(p->compute_bins, p->cache_bins, total_steps);
-}
-
-inline float get_preset_threshold(const std::string& preset) {
-    if (preset == "slow" || preset == "s" || preset == "S")
-        return 0.20f;
-    if (preset == "medium" || preset == "m" || preset == "M")
-        return 0.25f;
-    if (preset == "fast" || preset == "f" || preset == "F")
-        return 0.30f;
-    if (preset == "ultra" || preset == "u" || preset == "U")
-        return 0.34f;
-    return 0.08f;
-}
-
-inline int get_preset_warmup(const std::string& preset) {
-    if (preset == "slow" || preset == "s" || preset == "S")
-        return 8;
-    if (preset == "medium" || preset == "m" || preset == "M")
-        return 6;
-    if (preset == "fast" || preset == "f" || preset == "F")
-        return 6;
-    if (preset == "ultra" || preset == "u" || preset == "U")
-        return 4;
-    return 8;
-}
-
-inline int get_preset_Fn(const std::string& preset) {
-    if (preset == "slow" || preset == "s" || preset == "S")
-        return 8;
-    if (preset == "medium" || preset == "m" || preset == "M")
-        return 8;
-    if (preset == "fast" || preset == "f" || preset == "F")
-        return 6;
-    if (preset == "ultra" || preset == "u" || preset == "U")
-        return 4;
-    return 8;
-}
-
-inline int get_preset_Bn(const std::string& preset) {
-    (void)preset;
-    return 0;
-}
-
 inline void parse_dbcache_options(const std::string& opts, DBCacheConfig& cfg) {
     if (opts.empty())
         return;

src/clip.hpp

@@ -4,6 +4,7 @@
 #include "ggml_extend.hpp"
 #include "model.h"
 #include "tokenize_util.h"
+#include "vocab/vocab.h"
 
 /*================================================== CLIPTokenizer ===================================================*/
 
@@ -110,7 +111,7 @@ public:
         if (merges_utf8_str.size() > 0) {
             load_from_merges(merges_utf8_str);
         } else {
-            load_from_merges(ModelLoader::load_merges());
+            load_from_merges(load_clip_merges());
         }
         add_special_token("<|startoftext|>");
         add_special_token("<|endoftext|>");
@@ -479,9 +480,9 @@ public:
 
         x = fc1->forward(ctx, x);
         if (use_gelu) {
-            x = ggml_gelu_inplace(ctx->ggml_ctx, x);
+            x = ggml_ext_gelu(ctx->ggml_ctx, x, true);
         } else {
-            x = ggml_gelu_quick_inplace(ctx->ggml_ctx, x);
+            x = ggml_ext_gelu_quick(ctx->ggml_ctx, x, true);
         }
         x = fc2->forward(ctx, x);
         return x;
@@ -510,7 +511,7 @@ public:
         blocks["mlp"] = std::shared_ptr<GGMLBlock>(new CLIPMLP(d_model, intermediate_size));
     }
 
-    struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x, bool mask = true) {
+    struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x, struct ggml_tensor* mask = nullptr) {
         // x: [N, n_token, d_model]
         auto self_attn   = std::dynamic_pointer_cast<MultiheadAttention>(blocks["self_attn"]);
         auto layer_norm1 = std::dynamic_pointer_cast<LayerNorm>(blocks["layer_norm1"]);
@@ -542,8 +543,8 @@ public:
 
     struct ggml_tensor* forward(GGMLRunnerContext* ctx,
                                 struct ggml_tensor* x,
-                                int clip_skip = -1,
+                                struct ggml_tensor* mask = nullptr,
-                                bool mask     = true) {
+                                int clip_skip            = -1) {
         // x: [N, n_token, d_model]
         int layer_idx = n_layer - 1;
         // LOG_DEBUG("clip_skip %d", clip_skip);
@@ -741,6 +742,7 @@ public:
     struct ggml_tensor* forward(GGMLRunnerContext* ctx,
                                 struct ggml_tensor* input_ids,
                                 struct ggml_tensor* tkn_embeddings,
+                                struct ggml_tensor* mask = nullptr,
                                 size_t max_token_idx     = 0,
                                 bool return_pooled       = false,
                                 int clip_skip            = -1) {
@@ -750,7 +752,7 @@ public:
         auto final_layer_norm = std::dynamic_pointer_cast<LayerNorm>(blocks["final_layer_norm"]);
 
         auto x = embeddings->forward(ctx, input_ids, tkn_embeddings);  // [N, n_token, hidden_size]
-        x      = encoder->forward(ctx, x, return_pooled ? -1 : clip_skip, true);
+        x      = encoder->forward(ctx, x, mask, return_pooled ? -1 : clip_skip);
         if (return_pooled || with_final_ln) {
             x = final_layer_norm->forward(ctx, x);
         }
@@ -814,9 +816,10 @@ public:
 
         auto x = embeddings->forward(ctx, pixel_values);  // [N, num_positions, embed_dim]
         x      = pre_layernorm->forward(ctx, x);
-        x      = encoder->forward(ctx, x, clip_skip, false);
+        x      = encoder->forward(ctx, x, nullptr, clip_skip);
-        // print_ggml_tensor(x, true, "ClipVisionModel x: ");
+
         auto last_hidden_state = x;
+
         x = post_layernorm->forward(ctx, x);  // [N, n_token, hidden_size]
 
         GGML_ASSERT(x->ne[3] == 1);
@@ -905,6 +908,8 @@ public:
 struct CLIPTextModelRunner : public GGMLRunner {
     CLIPTextModel model;
 
+    std::vector<float> attention_mask_vec;
+
     CLIPTextModelRunner(ggml_backend_t backend,
                         bool offload_params_to_cpu,
                         const String2TensorStorage& tensor_storage_map,
@@ -938,6 +943,7 @@ struct CLIPTextModelRunner : public GGMLRunner {
     struct ggml_tensor* forward(GGMLRunnerContext* ctx,
                                 struct ggml_tensor* input_ids,
                                 struct ggml_tensor* embeddings,
+                                struct ggml_tensor* mask,
                                 size_t max_token_idx = 0,
                                 bool return_pooled   = false,
                                 int clip_skip        = -1) {
@@ -948,7 +954,7 @@ struct CLIPTextModelRunner : public GGMLRunner {
             input_ids = ggml_reshape_2d(ctx->ggml_ctx, input_ids, model.n_token, input_ids->ne[0] / model.n_token);
         }
 
-        return model.forward(ctx, input_ids, embeddings, max_token_idx, return_pooled, clip_skip);
+        return model.forward(ctx, input_ids, embeddings, mask, max_token_idx, return_pooled, clip_skip);
     }
 
     struct ggml_cgraph* build_graph(struct ggml_tensor* input_ids,
@@ -975,9 +981,23 @@ struct CLIPTextModelRunner : public GGMLRunner {
             embeddings = ggml_concat(compute_ctx, token_embed_weight, custom_embeddings, 1);
         }
 
+        int n_tokens = static_cast<int>(input_ids->ne[0]);
+        attention_mask_vec.resize(n_tokens * n_tokens);
+        for (int i0 = 0; i0 < n_tokens; i0++) {
+            for (int i1 = 0; i1 < n_tokens; i1++) {
+                float value = 0.f;
+                if (i0 > i1) {
+                    value = -INFINITY;
+                }
+                attention_mask_vec[i1 * n_tokens + i0] = value;
+            }
+        }
+        auto attention_mask = ggml_new_tensor_2d(compute_ctx, GGML_TYPE_F32, n_tokens, n_tokens);
+        set_backend_tensor_data(attention_mask, attention_mask_vec.data());
+
         auto runner_ctx = get_context();
 
-        struct ggml_tensor* hidden_states = forward(&runner_ctx, input_ids, embeddings, max_token_idx, return_pooled, clip_skip);
+        struct ggml_tensor* hidden_states = forward(&runner_ctx, input_ids, embeddings, attention_mask, max_token_idx, return_pooled, clip_skip);
 
         ggml_build_forward_expand(gf, hidden_states);
 

src/common_block.hpp

@@ -1,5 +1,5 @@
-#ifndef __COMMON_HPP__
+#ifndef __COMMON_BLOCK_HPP__
-#define __COMMON_HPP__
+#define __COMMON_BLOCK_HPP__
 
 #include "ggml_extend.hpp"
 
@@ -200,7 +200,7 @@ public:
 
         gate = ggml_cont(ctx->ggml_ctx, gate);
 
-        gate = ggml_gelu_inplace(ctx->ggml_ctx, gate);
+        gate = ggml_ext_gelu(ctx->ggml_ctx, gate, true);
 
         x = ggml_mul(ctx->ggml_ctx, x, gate);  // [ne3, ne2, ne1, dim_out]
 
@@ -220,7 +220,7 @@ public:
         auto proj = std::dynamic_pointer_cast<Linear>(blocks["proj"]);
 
         x = proj->forward(ctx, x);
-        x = ggml_gelu_inplace(ctx->ggml_ctx, x);
+        x = ggml_ext_gelu(ctx->ggml_ctx, x, true);
         return x;
     }
 };
@@ -317,7 +317,7 @@ public:
         auto k = to_k->forward(ctx, context);  // [N, n_context, inner_dim]
         auto v = to_v->forward(ctx, context);  // [N, n_context, inner_dim]
 
-        x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, n_head, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_token, inner_dim]
+        x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, n_head, nullptr, false, ctx->flash_attn_enabled);  // [N, n_token, inner_dim]
 
         x = to_out_0->forward(ctx, x);  // [N, n_token, query_dim]
         return x;
@@ -536,8 +536,8 @@ public:
         // image_only_indicator is always tensor([0.])
         float alpha = get_alpha();
         auto x      = ggml_add(ctx->ggml_ctx,
-                               ggml_scale(ctx->ggml_ctx, x_spatial, alpha),
+                               ggml_ext_scale(ctx->ggml_ctx, x_spatial, alpha),
-                               ggml_scale(ctx->ggml_ctx, x_temporal, 1.0f - alpha));
+                               ggml_ext_scale(ctx->ggml_ctx, x_temporal, 1.0f - alpha));
         return x;
     }
 };
@@ -590,4 +590,4 @@ public:
     }
 };
 
-#endif  // __COMMON_HPP__
+#endif  // __COMMON_BLOCK_HPP__

src/common_dit.hpp

@@ -0,0 +1,108 @@
+#ifndef __COMMON_DIT_HPP__
+#define __COMMON_DIT_HPP__
+
+#include "ggml_extend.hpp"
+
+namespace DiT {
+    ggml_tensor* patchify(ggml_context* ctx,
+                          ggml_tensor* x,
+                          int pw,
+                          int ph,
+                          bool patch_last = true) {
+        // x: [N, C, H, W]
+        // return: [N, h*w, C*ph*pw] if patch_last else [N, h*w, ph*pw*C]
+        int64_t N = x->ne[3];
+        int64_t C = x->ne[2];
+        int64_t H = x->ne[1];
+        int64_t W = x->ne[0];
+        int64_t h = H / ph;
+        int64_t w = W / pw;
+
+        GGML_ASSERT(h * ph == H && w * pw == W);
+
+        x = ggml_reshape_4d(ctx, x, pw, w, ph, h * C * N);     // [N*C*h, ph, w, pw]
+        x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N*C*h, w, ph, pw]
+        x = ggml_reshape_4d(ctx, x, pw * ph, w * h, C, N);     // [N, C, h*w, ph*pw]
+        if (patch_last) {
+            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N, h*w, C, ph*pw]
+            x = ggml_reshape_3d(ctx, x, pw * ph * C, w * h, N);    // [N, h*w, C*ph*pw]
+        } else {
+            x = ggml_cont(ctx, ggml_ext_torch_permute(ctx, x, 2, 0, 1, 3));  // [N, h*w, C, ph*pw]
+            x = ggml_reshape_3d(ctx, x, C * pw * ph, w * h, N);              // [N, h*w, ph*pw*C]
+        }
+        return x;
+    }
+
+    ggml_tensor* unpatchify(ggml_context* ctx,
+                            ggml_tensor* x,
+                            int64_t h,
+                            int64_t w,
+                            int ph,
+                            int pw,
+                            bool patch_last = true) {
+        // x: [N, h*w, C*ph*pw] if patch_last else [N, h*w, ph*pw*C]
+        // return: [N, C, H, W]
+        int64_t N = x->ne[2];
+        int64_t C = x->ne[0] / ph / pw;
+        int64_t H = h * ph;
+        int64_t W = w * pw;
+
+        GGML_ASSERT(C * ph * pw == x->ne[0]);
+
+        if (patch_last) {
+            x = ggml_reshape_4d(ctx, x, pw * ph, C, w * h, N);     // [N, h*w, C, ph*pw]
+            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N, C, h*w, ph*pw]
+        } else {
+            x = ggml_reshape_4d(ctx, x, C, pw * ph, w * h, N);     // [N, h*w, ph*pw, C]
+            x = ggml_cont(ctx, ggml_permute(ctx, x, 2, 0, 1, 3));  // [N, C, h*w, ph*pw]
+        }
+
+        x = ggml_reshape_4d(ctx, x, pw, ph, w, h * C * N);     // [N*C*h, w, ph, pw]
+        x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N*C*h, ph, w, pw]
+        x = ggml_reshape_4d(ctx, x, W, H, C, N);               // [N, C, h*ph, w*pw]
+
+        return x;
+    }
+
+    ggml_tensor* pad_to_patch_size(GGMLRunnerContext* ctx,
+                                   ggml_tensor* x,
+                                   int ph,
+                                   int pw) {
+        int64_t W = x->ne[0];
+        int64_t H = x->ne[1];
+
+        int pad_h = (ph - H % ph) % ph;
+        int pad_w = (pw - W % pw) % pw;
+        x         = ggml_ext_pad(ctx->ggml_ctx, x, pad_w, pad_h, 0, 0, ctx->circular_x_enabled, ctx->circular_y_enabled);
+        return x;
+    }
+
+    ggml_tensor* pad_and_patchify(GGMLRunnerContext* ctx,
+                                  ggml_tensor* x,
+                                  int ph,
+                                  int pw,
+                                  bool patch_last = true) {
+        x = pad_to_patch_size(ctx, x, ph, pw);
+        x = patchify(ctx->ggml_ctx, x, ph, pw, patch_last);
+        return x;
+    }
+
+    ggml_tensor* unpatchify_and_crop(ggml_context* ctx,
+                                     ggml_tensor* x,
+                                     int64_t H,
+                                     int64_t W,
+                                     int ph,
+                                     int pw,
+                                     bool patch_last = true) {
+        int pad_h = (ph - H % ph) % ph;
+        int pad_w = (pw - W % pw) % pw;
+        int64_t h = ((H + pad_h) / ph);
+        int64_t w = ((W + pad_w) / pw);
+        x         = unpatchify(ctx, x, h, w, ph, pw, patch_last);  // [N, C, H + pad_h, W + pad_w]
+        x         = ggml_ext_slice(ctx, x, 1, 0, H);               // [N, C, H, W + pad_w]
+        x         = ggml_ext_slice(ctx, x, 0, 0, W);               // [N, C, H, W]
+        return x;
+    }
+}  // namespace DiT
+
+#endif  // __COMMON_DIT_HPP__

src/conditioner.hpp

@@ -10,9 +10,14 @@ struct SDCondition {
     struct ggml_tensor* c_vector    = nullptr;  // aka y
     struct ggml_tensor* c_concat    = nullptr;
 
+    std::vector<struct ggml_tensor*> extra_c_crossattns;
+
     SDCondition() = default;
-    SDCondition(struct ggml_tensor* c_crossattn, struct ggml_tensor* c_vector, struct ggml_tensor* c_concat)
+    SDCondition(struct ggml_tensor* c_crossattn,
-        : c_crossattn(c_crossattn), c_vector(c_vector), c_concat(c_concat) {}
+                struct ggml_tensor* c_vector,
+                struct ggml_tensor* c_concat,
+                const std::vector<struct ggml_tensor*>& extra_c_crossattns = {})
+        : c_crossattn(c_crossattn), c_vector(c_vector), c_concat(c_concat), extra_c_crossattns(extra_c_crossattns) {}
 };
 
 struct ConditionerParams {
@@ -34,6 +39,7 @@ struct Conditioner {
     virtual void free_params_buffer()                                                      = 0;
     virtual void get_param_tensors(std::map<std::string, struct ggml_tensor*>& tensors)    = 0;
     virtual size_t get_params_buffer_size()                                                = 0;
+    virtual void set_flash_attention_enabled(bool enabled)                                 = 0;
     virtual void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) {}
     virtual std::tuple<SDCondition, std::vector<bool>> get_learned_condition_with_trigger(ggml_context* work_ctx,
                                                                                           int n_threads,
@@ -115,6 +121,13 @@ struct FrozenCLIPEmbedderWithCustomWords : public Conditioner {
         return buffer_size;
     }
 
+    void set_flash_attention_enabled(bool enabled) override {
+        text_model->set_flash_attention_enabled(enabled);
+        if (sd_version_is_sdxl(version)) {
+            text_model2->set_flash_attention_enabled(enabled);
+        }
+    }
+
     void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) override {
         text_model->set_weight_adapter(adapter);
         if (sd_version_is_sdxl(version)) {
@@ -783,6 +796,18 @@ struct SD3CLIPEmbedder : public Conditioner {
         return buffer_size;
     }
 
+    void set_flash_attention_enabled(bool enabled) override {
+        if (clip_l) {
+            clip_l->set_flash_attention_enabled(enabled);
+        }
+        if (clip_g) {
+            clip_g->set_flash_attention_enabled(enabled);
+        }
+        if (t5) {
+            t5->set_flash_attention_enabled(enabled);
+        }
+    }
+
     void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) override {
         if (clip_l) {
             clip_l->set_weight_adapter(adapter);
@@ -1191,6 +1216,15 @@ struct FluxCLIPEmbedder : public Conditioner {
         return buffer_size;
     }
 
+    void set_flash_attention_enabled(bool enabled) override {
+        if (clip_l) {
+            clip_l->set_flash_attention_enabled(enabled);
+        }
+        if (t5) {
+            t5->set_flash_attention_enabled(enabled);
+        }
+    }
+
     void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) {
         if (clip_l) {
             clip_l->set_weight_adapter(adapter);
@@ -1440,6 +1474,12 @@ struct T5CLIPEmbedder : public Conditioner {
         return buffer_size;
     }
 
+    void set_flash_attention_enabled(bool enabled) override {
+        if (t5) {
+            t5->set_flash_attention_enabled(enabled);
+        }
+    }
+
     void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) override {
         if (t5) {
             t5->set_weight_adapter(adapter);
@@ -1601,6 +1641,142 @@ struct T5CLIPEmbedder : public Conditioner {
     }
 };
 
+struct AnimaConditioner : public Conditioner {
+    std::shared_ptr<LLM::BPETokenizer> qwen_tokenizer;
+    T5UniGramTokenizer t5_tokenizer;
+    std::shared_ptr<LLM::LLMRunner> llm;
+
+    AnimaConditioner(ggml_backend_t backend,
+                     bool offload_params_to_cpu,
+                     const String2TensorStorage& tensor_storage_map = {}) {
+        qwen_tokenizer = std::make_shared<LLM::Qwen2Tokenizer>();
+        llm            = std::make_shared<LLM::LLMRunner>(LLM::LLMArch::QWEN3,
+                                               backend,
+                                               offload_params_to_cpu,
+                                               tensor_storage_map,
+                                               "text_encoders.llm",
+                                               false);
+    }
+
+    void get_param_tensors(std::map<std::string, struct ggml_tensor*>& tensors) override {
+        llm->get_param_tensors(tensors, "text_encoders.llm");
+    }
+
+    void alloc_params_buffer() override {
+        llm->alloc_params_buffer();
+    }
+
+    void free_params_buffer() override {
+        llm->free_params_buffer();
+    }
+
+    size_t get_params_buffer_size() override {
+        return llm->get_params_buffer_size();
+    }
+
+    void set_flash_attention_enabled(bool enabled) override {
+        llm->set_flash_attention_enabled(enabled);
+    }
+
+    void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) override {
+        llm->set_weight_adapter(adapter);
+    }
+
+    std::tuple<std::vector<int>, std::vector<float>, std::vector<int>, std::vector<float>> tokenize(std::string text) {
+        auto parsed_attention = parse_prompt_attention(text);
+
+        {
+            std::stringstream ss;
+            ss << "[";
+            for (const auto& item : parsed_attention) {
+                ss << "['" << item.first << "', " << item.second << "], ";
+            }
+            ss << "]";
+            LOG_DEBUG("parse '%s' to %s", text.c_str(), ss.str().c_str());
+        }
+
+        std::vector<int> qwen_tokens;
+        std::vector<float> qwen_weights;
+        std::vector<int> t5_tokens;
+        std::vector<float> t5_weights;
+
+        for (const auto& item : parsed_attention) {
+            const std::string& curr_text = item.first;
+            std::vector<int> curr_tokens = qwen_tokenizer->tokenize(curr_text, nullptr);
+            qwen_tokens.insert(qwen_tokens.end(), curr_tokens.begin(), curr_tokens.end());
+            // Anima uses uniform Qwen token weights.
+            qwen_weights.insert(qwen_weights.end(), curr_tokens.size(), 1.f);
+        }
+        if (qwen_tokens.empty()) {
+            qwen_tokens.push_back(151643);  // qwen3 pad token
+            qwen_weights.push_back(1.f);
+        }
+
+        for (const auto& item : parsed_attention) {
+            const std::string& curr_text = item.first;
+            float curr_weight            = item.second;
+            std::vector<int> curr_tokens = t5_tokenizer.Encode(curr_text, true);
+            t5_tokens.insert(t5_tokens.end(), curr_tokens.begin(), curr_tokens.end());
+            t5_weights.insert(t5_weights.end(), curr_tokens.size(), curr_weight);
+        }
+
+        return {qwen_tokens, qwen_weights, t5_tokens, t5_weights};
+    }
+
+    SDCondition get_learned_condition(ggml_context* work_ctx,
+                                      int n_threads,
+                                      const ConditionerParams& conditioner_params) override {
+        int64_t t0 = ggml_time_ms();
+
+        auto tokenized     = tokenize(conditioner_params.text);
+        auto& qwen_tokens  = std::get<0>(tokenized);
+        auto& qwen_weights = std::get<1>(tokenized);
+        auto& t5_tokens    = std::get<2>(tokenized);
+        auto& t5_weights   = std::get<3>(tokenized);
+
+        auto input_ids = vector_to_ggml_tensor_i32(work_ctx, qwen_tokens);
+
+        struct ggml_tensor* hidden_states = nullptr;  // [N, n_token, 1024]
+        llm->compute(n_threads,
+                     input_ids,
+                     nullptr,
+                     {},
+                     {},
+                     &hidden_states,
+                     work_ctx);
+
+        {
+            auto tensor         = hidden_states;
+            float original_mean = ggml_ext_tensor_mean(tensor);
+            for (int i2 = 0; i2 < tensor->ne[2]; i2++) {
+                for (int i1 = 0; i1 < tensor->ne[1]; i1++) {
+                    for (int i0 = 0; i0 < tensor->ne[0]; i0++) {
+                        float value = ggml_ext_tensor_get_f32(tensor, i0, i1, i2);
+                        value *= qwen_weights[i1];
+                        ggml_ext_tensor_set_f32(tensor, value, i0, i1, i2);
+                    }
+                }
+            }
+            float new_mean = ggml_ext_tensor_mean(tensor);
+            if (new_mean != 0.f) {
+                ggml_ext_tensor_scale_inplace(tensor, (original_mean / new_mean));
+            }
+        }
+
+        struct ggml_tensor* t5_ids_tensor    = nullptr;
+        struct ggml_tensor* t5_weight_tensor = nullptr;
+        if (!t5_tokens.empty()) {
+            t5_ids_tensor    = vector_to_ggml_tensor_i32(work_ctx, t5_tokens);
+            t5_weight_tensor = vector_to_ggml_tensor(work_ctx, t5_weights);
+        }
+
+        int64_t t1 = ggml_time_ms();
+        LOG_DEBUG("computing condition graph completed, taking %" PRId64 " ms", t1 - t0);
+
+        return {hidden_states, t5_weight_tensor, t5_ids_tensor};
+    }
+};
+
 struct LLMEmbedder : public Conditioner {
     SDVersion version;
     std::shared_ptr<LLM::BPETokenizer> tokenizer;
@@ -1650,6 +1826,10 @@ struct LLMEmbedder : public Conditioner {
         return buffer_size;
     }
 
+    void set_flash_attention_enabled(bool enabled) override {
+        llm->set_flash_attention_enabled(enabled);
+    }
+
     void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) override {
         if (llm) {
             llm->set_weight_adapter(adapter);
@@ -1657,10 +1837,11 @@ struct LLMEmbedder : public Conditioner {
     }
 
     std::tuple<std::vector<int>, std::vector<float>> tokenize(std::string text,
-                                                              std::pair<int, int> attn_range,
+                                                              const std::pair<int, int>& attn_range,
                                                               size_t max_length = 0,
                                                               bool padding      = false) {
         std::vector<std::pair<std::string, float>> parsed_attention;
+        if (attn_range.first >= 0 && attn_range.second > 0) {
             parsed_attention.emplace_back(text.substr(0, attn_range.first), 1.f);
             if (attn_range.second - attn_range.first > 0) {
                 auto new_parsed_attention = parse_prompt_attention(text.substr(attn_range.first, attn_range.second - attn_range.first));
@@ -1669,6 +1850,10 @@ struct LLMEmbedder : public Conditioner {
                                         new_parsed_attention.end());
             }
             parsed_attention.emplace_back(text.substr(attn_range.second), 1.f);
+        } else {
+            parsed_attention.emplace_back(text, 1.f);
+        }
+
         {
             std::stringstream ss;
             ss << "[";
@@ -1699,19 +1884,110 @@ struct LLMEmbedder : public Conditioner {
         return {tokens, weights};
     }
 
+    ggml_tensor* encode_prompt(ggml_context* work_ctx,
+                               int n_threads,
+                               const std::string prompt,
+                               const std::pair<int, int>& prompt_attn_range,
+                               int max_length,
+                               int min_length,
+                               std::vector<std::pair<int, ggml_tensor*>> image_embeds,
+                               const std::set<int>& out_layers,
+                               int prompt_template_encode_start_idx) {
+        auto tokens_and_weights = tokenize(prompt, prompt_attn_range);
+        auto& tokens            = std::get<0>(tokens_and_weights);
+        auto& weights           = std::get<1>(tokens_and_weights);
+        std::vector<float> mask;
+
+        if (max_length > 0 && tokens.size() < max_length) {
+            mask.insert(mask.end(), tokens.size(), 1.f);
+            mask.insert(mask.end(), max_length - tokens.size(), 0.f);
+            tokenizer->pad_tokens(tokens, weights, max_length, true);
+        }
+
+        struct ggml_tensor* hidden_states = nullptr;  // [N, n_token, hidden_size]
+
+        auto input_ids = vector_to_ggml_tensor_i32(work_ctx, tokens);
+
+        ggml_tensor* attention_mask = nullptr;
+        if (!mask.empty()) {
+            attention_mask = ggml_new_tensor_2d(work_ctx, GGML_TYPE_F32, mask.size(), mask.size());
+            ggml_ext_tensor_iter(attention_mask, [&](ggml_tensor* attention_mask, int64_t i0, int64_t i1, int64_t i2, int64_t i3) {
+                float value = 0.f;
+                if (mask[i0] == 0.f) {
+                    value = -INFINITY;
+                } else if (i0 > i1) {
+                    value = -INFINITY;
+                }
+                ggml_ext_tensor_set_f32(attention_mask, value, i0, i1, i2, i3);
+            });
+        }
+
+        llm->compute(n_threads,
+                     input_ids,
+                     attention_mask,
+                     image_embeds,
+                     out_layers,
+                     &hidden_states,
+                     work_ctx);
+        {
+            auto tensor         = hidden_states;
+            float original_mean = ggml_ext_tensor_mean(tensor);
+            for (int i2 = 0; i2 < tensor->ne[2]; i2++) {
+                for (int i1 = 0; i1 < tensor->ne[1]; i1++) {
+                    for (int i0 = 0; i0 < tensor->ne[0]; i0++) {
+                        float value = ggml_ext_tensor_get_f32(tensor, i0, i1, i2);
+                        value *= weights[i1];
+                        ggml_ext_tensor_set_f32(tensor, value, i0, i1, i2);
+                    }
+                }
+            }
+            float new_mean = ggml_ext_tensor_mean(tensor);
+            ggml_ext_tensor_scale_inplace(tensor, (original_mean / new_mean));
+        }
+
+        GGML_ASSERT(hidden_states->ne[1] > prompt_template_encode_start_idx);
+
+        int64_t zero_pad_len = 0;
+        if (min_length > 0) {
+            if (hidden_states->ne[1] - prompt_template_encode_start_idx < min_length) {
+                zero_pad_len = min_length - hidden_states->ne[1] + prompt_template_encode_start_idx;
+            }
+        }
+
+        ggml_tensor* new_hidden_states = ggml_new_tensor_3d(work_ctx,
+                                                            GGML_TYPE_F32,
+                                                            hidden_states->ne[0],
+                                                            hidden_states->ne[1] - prompt_template_encode_start_idx + zero_pad_len,
+                                                            hidden_states->ne[2]);
+
+        ggml_ext_tensor_iter(new_hidden_states, [&](ggml_tensor* new_hidden_states, int64_t i0, int64_t i1, int64_t i2, int64_t i3) {
+            float value = 0.f;
+            if (i1 + prompt_template_encode_start_idx < hidden_states->ne[1]) {
+                value = ggml_ext_tensor_get_f32(hidden_states, i0, i1 + prompt_template_encode_start_idx, i2, i3);
+            }
+            ggml_ext_tensor_set_f32(new_hidden_states, value, i0, i1, i2, i3);
+        });
+
+        return new_hidden_states;
+    }
+
     SDCondition get_learned_condition(ggml_context* work_ctx,
                                       int n_threads,
                                       const ConditionerParams& conditioner_params) override {
         std::string prompt;
-        std::vector<std::pair<int, ggml_tensor*>> image_embeds;
         std::pair<int, int> prompt_attn_range;
+        std::vector<std::string> extra_prompts;
+        std::vector<std::pair<int, int>> extra_prompts_attn_range;
+        std::vector<std::pair<int, ggml_tensor*>> image_embeds;
         int prompt_template_encode_start_idx = 34;
-        int max_length                       = 0;
+        int max_length                       = 0;  // pad tokens
+        int min_length                       = 0;  // zero pad hidden_states
         std::set<int> out_layers;
-        std::vector<int> tokens;
+
-        std::vector<float> weights;
+        int64_t t0 = ggml_time_ms();
-        std::vector<float> mask;
+
-        if (llm->enable_vision && conditioner_params.ref_images.size() > 0) {
+        if (sd_version_is_qwen_image(version)) {
+            if (llm->enable_vision && !conditioner_params.ref_images.empty()) {
                 LOG_INFO("QwenImageEditPlusPipeline");
                 prompt_template_encode_start_idx = 64;
                 int image_embed_idx              = 64 + 6;
@@ -1774,8 +2050,20 @@ struct LLMEmbedder : public Conditioner {
                 prompt_attn_range.second = static_cast<int>(prompt.size());
 
                 prompt += "<|im_end|>\n<|im_start|>assistant\n";
+            } else {
+                prompt_template_encode_start_idx = 34;
+
+                prompt = "<|im_start|>system\nDescribe the image by detailing the color, shape, size, texture, quantity, text, spatial relationships of the objects and background:<|im_end|>\n<|im_start|>user\n";
+
+                prompt_attn_range.first = static_cast<int>(prompt.size());
+                prompt += conditioner_params.text;
+                prompt_attn_range.second = static_cast<int>(prompt.size());
+
+                prompt += "<|im_end|>\n<|im_start|>assistant\n";
+            }
         } else if (version == VERSION_FLUX2) {
             prompt_template_encode_start_idx = 0;
+            min_length                       = 512;
             out_layers                       = {10, 20, 30};
 
             prompt = "[SYSTEM_PROMPT]You are an AI that reasons about image descriptions. You give structured responses focusing on object relationships, object\nattribution and actions without speculation.[/SYSTEM_PROMPT][INST]";
@@ -1789,6 +2077,15 @@ struct LLMEmbedder : public Conditioner {
             prompt_template_encode_start_idx = 0;
             out_layers                       = {35};  // -2
 
+            if (!conditioner_params.ref_images.empty()) {
+                LOG_INFO("ZImageOmniPipeline");
+                prompt = "<|im_start|>user\n<|vision_start|>";
+                for (int i = 0; i < conditioner_params.ref_images.size() - 1; i++) {
+                    extra_prompts.push_back("<|vision_end|><|vision_start|>");
+                }
+                extra_prompts.push_back("<|vision_end|>" + conditioner_params.text + "<|im_end|>\n<|im_start|>assistant\n<|vision_start|>");
+                extra_prompts.push_back("<|vision_end|><|im_end|>");
+            } else {
                 prompt = "<|im_start|>user\n";
 
                 prompt_attn_range.first = static_cast<int>(prompt.size());
@@ -1796,6 +2093,7 @@ struct LLMEmbedder : public Conditioner {
                 prompt_attn_range.second = static_cast<int>(prompt.size());
 
                 prompt += "<|im_end|>\n<|im_start|>assistant\n";
+            }
         } else if (version == VERSION_FLUX2_KLEIN) {
             prompt_template_encode_start_idx = 0;
             max_length                       = 512;
@@ -1808,16 +2106,6 @@ struct LLMEmbedder : public Conditioner {
             prompt_attn_range.second = static_cast<int>(prompt.size());
 
             prompt += "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n";
-
-            auto tokens_and_weights = tokenize(prompt, prompt_attn_range, 0, false);
-            tokens                  = std::get<0>(tokens_and_weights);
-            weights                 = std::get<1>(tokens_and_weights);
-
-            mask.insert(mask.end(), tokens.size(), 1.f);
-            if (tokens.size() < max_length) {
-                mask.insert(mask.end(), max_length - tokens.size(), 0.f);
-                tokenizer->pad_tokens(tokens, weights, max_length, true);
-            }
         } else if (version == VERSION_OVIS_IMAGE) {
             prompt_template_encode_start_idx = 28;
             max_length                       = prompt_template_encode_start_idx + 256;
@@ -1830,98 +2118,36 @@ struct LLMEmbedder : public Conditioner {
 
             prompt += "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n";
         } else {
-            prompt_template_encode_start_idx = 34;
+            GGML_ABORT("unknown version %d", version);
-
-            prompt = "<|im_start|>system\nDescribe the image by detailing the color, shape, size, texture, quantity, text, spatial relationships of the objects and background:<|im_end|>\n<|im_start|>user\n";
-
-            prompt_attn_range.first = static_cast<int>(prompt.size());
-            prompt += conditioner_params.text;
-            prompt_attn_range.second = static_cast<int>(prompt.size());
-
-            prompt += "<|im_end|>\n<|im_start|>assistant\n";
         }
 
-        if (tokens.empty()) {
+        auto hidden_states = encode_prompt(work_ctx,
-            auto tokens_and_weights = tokenize(prompt, prompt_attn_range, max_length, max_length > 0);
+                                           n_threads,
-            tokens                  = std::get<0>(tokens_and_weights);
+                                           prompt,
-            weights                 = std::get<1>(tokens_and_weights);
+                                           prompt_attn_range,
-        }
+                                           max_length,
-
+                                           min_length,
-        int64_t t0                        = ggml_time_ms();
-        struct ggml_tensor* hidden_states = nullptr;  // [N, n_token, 3584]
-
-        auto input_ids = vector_to_ggml_tensor_i32(work_ctx, tokens);
-
-        ggml_tensor* attention_mask = nullptr;
-        if (!mask.empty()) {
-            attention_mask = ggml_new_tensor_2d(work_ctx, GGML_TYPE_F32, mask.size(), mask.size());
-            ggml_ext_tensor_iter(attention_mask, [&](ggml_tensor* attention_mask, int64_t i0, int64_t i1, int64_t i2, int64_t i3) {
-                float value = 0.f;
-                if (mask[i0] == 0.f) {
-                    value = -INFINITY;
-                } else if (i0 > i1) {
-                    value = -INFINITY;
-                }
-                ggml_ext_tensor_set_f32(attention_mask, value, i0, i1, i2, i3);
-            });
-        }
-
-        llm->compute(n_threads,
-                     input_ids,
-                     attention_mask,
                                            image_embeds,
                                            out_layers,
-                     &hidden_states,
+                                           prompt_template_encode_start_idx);
-                     work_ctx);
-        {
-            auto tensor         = hidden_states;
-            float original_mean = ggml_ext_tensor_mean(tensor);
-            for (int i2 = 0; i2 < tensor->ne[2]; i2++) {
-                for (int i1 = 0; i1 < tensor->ne[1]; i1++) {
-                    for (int i0 = 0; i0 < tensor->ne[0]; i0++) {
-                        float value = ggml_ext_tensor_get_f32(tensor, i0, i1, i2);
-                        value *= weights[i1];
-                        ggml_ext_tensor_set_f32(tensor, value, i0, i1, i2);
-                    }
-                }
-            }
-            float new_mean = ggml_ext_tensor_mean(tensor);
-            ggml_ext_tensor_scale_inplace(tensor, (original_mean / new_mean));
-        }
 
-        GGML_ASSERT(hidden_states->ne[1] > prompt_template_encode_start_idx);
+        std::vector<ggml_tensor*> extra_hidden_states_vec;
-
+        for (int i = 0; i < extra_prompts.size(); i++) {
-        int64_t min_length = 0;
+            auto extra_hidden_states = encode_prompt(work_ctx,
-        if (version == VERSION_FLUX2) {
+                                                     n_threads,
-            min_length = 512;
+                                                     extra_prompts[i],
+                                                     extra_prompts_attn_range[i],
+                                                     max_length,
+                                                     min_length,
+                                                     image_embeds,
+                                                     out_layers,
+                                                     prompt_template_encode_start_idx);
+            extra_hidden_states_vec.push_back(extra_hidden_states);
         }
 
-        int64_t zero_pad_len = 0;
-        if (min_length > 0) {
-            if (hidden_states->ne[1] - prompt_template_encode_start_idx < min_length) {
-                zero_pad_len = min_length - hidden_states->ne[1] + prompt_template_encode_start_idx;
-            }
-        }
-
-        ggml_tensor* new_hidden_states = ggml_new_tensor_3d(work_ctx,
-                                                            GGML_TYPE_F32,
-                                                            hidden_states->ne[0],
-                                                            hidden_states->ne[1] - prompt_template_encode_start_idx + zero_pad_len,
-                                                            hidden_states->ne[2]);
-
-        ggml_ext_tensor_iter(new_hidden_states, [&](ggml_tensor* new_hidden_states, int64_t i0, int64_t i1, int64_t i2, int64_t i3) {
-            float value = 0.f;
-            if (i1 + prompt_template_encode_start_idx < hidden_states->ne[1]) {
-                value = ggml_ext_tensor_get_f32(hidden_states, i0, i1 + prompt_template_encode_start_idx, i2, i3);
-            }
-            ggml_ext_tensor_set_f32(new_hidden_states, value, i0, i1, i2, i3);
-        });
-
-        // print_ggml_tensor(new_hidden_states);
-
         int64_t t1 = ggml_time_ms();
         LOG_DEBUG("computing condition graph completed, taking %" PRId64 " ms", t1 - t0);
-        return {new_hidden_states, nullptr, nullptr};
+        return {hidden_states, nullptr, nullptr, extra_hidden_states_vec};
     }
 };
 

src/control.hpp

@@ -1,8 +1,7 @@
 #ifndef __CONTROL_HPP__
 #define __CONTROL_HPP__
 
-#include "common.hpp"
+#include "common_block.hpp"
-#include "ggml_extend.hpp"
 #include "model.h"
 
 #define CONTROL_NET_GRAPH_SIZE 1536

src/denoiser.hpp

@@ -1,6 +1,8 @@
 #ifndef __DENOISER_HPP__
 #define __DENOISER_HPP__
 
+#include <cmath>
+
 #include "ggml_extend.hpp"
 #include "gits_noise.inl"
 
@@ -351,6 +353,95 @@ struct SmoothStepScheduler : SigmaScheduler {
     }
 };
 
+struct BongTangentScheduler : SigmaScheduler {
+    static constexpr float kPi = 3.14159265358979323846f;
+
+    static std::vector<float> get_bong_tangent_sigmas(int steps, float slope, float pivot, float start, float end) {
+        std::vector<float> sigmas;
+        if (steps <= 0) {
+            return sigmas;
+        }
+
+        float smax   = ((2.0f / kPi) * atanf(-slope * (0.0f - pivot)) + 1.0f) * 0.5f;
+        float smin   = ((2.0f / kPi) * atanf(-slope * ((float)(steps - 1) - pivot)) + 1.0f) * 0.5f;
+        float srange = smax - smin;
+        float sscale = start - end;
+
+        sigmas.reserve(steps);
+
+        if (fabsf(srange) < 1e-8f) {
+            if (steps == 1) {
+                sigmas.push_back(start);
+                return sigmas;
+            }
+            for (int i = 0; i < steps; ++i) {
+                float t = (float)i / (float)(steps - 1);
+                sigmas.push_back(start + (end - start) * t);
+            }
+            return sigmas;
+        }
+
+        float inv_srange = 1.0f / srange;
+        for (int x = 0; x < steps; ++x) {
+            float v     = ((2.0f / kPi) * atanf(-slope * ((float)x - pivot)) + 1.0f) * 0.5f;
+            float sigma = ((v - smin) * inv_srange) * sscale + end;
+            sigmas.push_back(sigma);
+        }
+
+        return sigmas;
+    }
+
+    std::vector<float> get_sigmas(uint32_t n, float sigma_min, float sigma_max, t_to_sigma_t /*t_to_sigma*/) override {
+        std::vector<float> result;
+        if (n == 0) {
+            return result;
+        }
+
+        float start  = sigma_max;
+        float end    = sigma_min;
+        float middle = sigma_min + (sigma_max - sigma_min) * 0.5f;
+
+        float pivot_1 = 0.6f;
+        float pivot_2 = 0.6f;
+        float slope_1 = 0.2f;
+        float slope_2 = 0.2f;
+
+        int steps     = static_cast<int>(n) + 2;
+        int midpoint  = static_cast<int>(((float)steps * pivot_1 + (float)steps * pivot_2) * 0.5f);
+        int pivot_1_i = static_cast<int>((float)steps * pivot_1);
+        int pivot_2_i = static_cast<int>((float)steps * pivot_2);
+
+        float slope_scale = (float)steps / 40.0f;
+        slope_1           = slope_1 / slope_scale;
+        slope_2           = slope_2 / slope_scale;
+
+        int stage_2_len = steps - midpoint;
+        int stage_1_len = steps - stage_2_len;
+
+        std::vector<float> sigmas_1 = get_bong_tangent_sigmas(stage_1_len, slope_1, (float)pivot_1_i, start, middle);
+        std::vector<float> sigmas_2 = get_bong_tangent_sigmas(stage_2_len, slope_2, (float)(pivot_2_i - stage_1_len), middle, end);
+
+        if (!sigmas_1.empty()) {
+            sigmas_1.pop_back();
+        }
+
+        result.reserve(n + 1);
+        result.insert(result.end(), sigmas_1.begin(), sigmas_1.end());
+        result.insert(result.end(), sigmas_2.begin(), sigmas_2.end());
+
+        if (result.size() < n + 1) {
+            while (result.size() < n + 1) {
+                result.push_back(end);
+            }
+        } else if (result.size() > n + 1) {
+            result.resize(n + 1);
+        }
+
+        result[n] = 0.0f;
+        return result;
+    }
+};
+
 struct KLOptimalScheduler : SigmaScheduler {
     std::vector<float> get_sigmas(uint32_t n, float sigma_min, float sigma_max, t_to_sigma_t t_to_sigma) override {
         std::vector<float> sigmas;
@@ -431,6 +522,10 @@ struct Denoiser {
                 LOG_INFO("get_sigmas with SmoothStep scheduler");
                 scheduler = std::make_shared<SmoothStepScheduler>();
                 break;
+            case BONG_TANGENT_SCHEDULER:
+                LOG_INFO("get_sigmas with bong_tangent scheduler");
+                scheduler = std::make_shared<BongTangentScheduler>();
+                break;
             case KL_OPTIMAL_SCHEDULER:
                 LOG_INFO("get_sigmas with KL Optimal scheduler");
                 scheduler = std::make_shared<KLOptimalScheduler>();
@@ -562,9 +657,8 @@ struct DiscreteFlowDenoiser : public Denoiser {
 
     float sigma_data = 1.0f;
 
-    DiscreteFlowDenoiser(float shift = 3.0f)
+    DiscreteFlowDenoiser(float shift = 3.0f) {
-        : shift(shift) {
+        set_shift(shift);
-        set_parameters();
     }
 
     void set_parameters() {
@@ -573,6 +667,11 @@ struct DiscreteFlowDenoiser : public Denoiser {
         }
     }
 
+    void set_shift(float shift) {
+        this->shift = shift;
+        set_parameters();
+    }
+
     float sigma_min() override {
         return sigmas[0];
     }
@@ -615,34 +714,8 @@ float flux_time_shift(float mu, float sigma, float t) {
     return ::expf(mu) / (::expf(mu) + ::powf((1.0f / t - 1.0f), sigma));
 }
 
-struct FluxFlowDenoiser : public Denoiser {
+struct FluxFlowDenoiser : public DiscreteFlowDenoiser {
-    float sigmas[TIMESTEPS];
+    FluxFlowDenoiser() = default;
-    float shift = 1.15f;
-
-    float sigma_data = 1.0f;
-
-    FluxFlowDenoiser(float shift = 1.15f) {
-        set_parameters(shift);
-    }
-
-    void set_shift(float shift) {
-        this->shift = shift;
-    }
-
-    void set_parameters(float shift) {
-        set_shift(shift);
-        for (int i = 0; i < TIMESTEPS; i++) {
-            sigmas[i] = t_to_sigma(static_cast<float>(i));
-        }
-    }
-
-    float sigma_min() override {
-        return sigmas[0];
-    }
-
-    float sigma_max() override {
-        return sigmas[TIMESTEPS - 1];
-    }
 
     float sigma_to_t(float sigma) override {
         return sigma;
@@ -652,26 +725,6 @@ struct FluxFlowDenoiser : public Denoiser {
         t = t + 1;
         return flux_time_shift(shift, 1.0f, t / TIMESTEPS);
     }
-
-    std::vector<float> get_scalings(float sigma) override {
-        float c_skip = 1.0f;
-        float c_out  = -sigma;
-        float c_in   = 1.0f;
-        return {c_skip, c_out, c_in};
-    }
-
-    // this function will modify noise/latent
-    ggml_tensor* noise_scaling(float sigma, ggml_tensor* noise, ggml_tensor* latent) override {
-        ggml_ext_tensor_scale_inplace(noise, sigma);
-        ggml_ext_tensor_scale_inplace(latent, 1.0f - sigma);
-        ggml_ext_tensor_add_inplace(latent, noise);
-        return latent;
-    }
-
-    ggml_tensor* inverse_noise_scaling(float sigma, ggml_tensor* latent) override {
-        ggml_ext_tensor_scale_inplace(latent, 1.0f / (1.0f - sigma));
-        return latent;
-    }
 };
 
 struct Flux2FlowDenoiser : public FluxFlowDenoiser {
@@ -1634,6 +1687,216 @@ static bool sample_k_diffusion(sample_method_t method,
                 }
             }
         } break;
+        case RES_MULTISTEP_SAMPLE_METHOD:  // Res Multistep sampler
+        {
+            struct ggml_tensor* noise        = ggml_dup_tensor(work_ctx, x);
+            struct ggml_tensor* old_denoised = ggml_dup_tensor(work_ctx, x);
+
+            bool have_old_sigma  = false;
+            float old_sigma_down = 0.0f;
+
+            auto t_fn     = [](float sigma) -> float { return -logf(sigma); };
+            auto sigma_fn = [](float t) -> float { return expf(-t); };
+            auto phi1_fn  = [](float t) -> float {
+                if (fabsf(t) < 1e-6f) {
+                    return 1.0f + t * 0.5f + (t * t) / 6.0f;
+                }
+                return (expf(t) - 1.0f) / t;
+            };
+            auto phi2_fn = [&](float t) -> float {
+                if (fabsf(t) < 1e-6f) {
+                    return 0.5f + t / 6.0f + (t * t) / 24.0f;
+                }
+                float phi1_val = phi1_fn(t);
+                return (phi1_val - 1.0f) / t;
+            };
+
+            for (int i = 0; i < steps; i++) {
+                ggml_tensor* denoised = model(x, sigmas[i], i + 1);
+                if (denoised == nullptr) {
+                    return false;
+                }
+
+                float sigma_from = sigmas[i];
+                float sigma_to   = sigmas[i + 1];
+                float sigma_up   = 0.0f;
+                float sigma_down = sigma_to;
+
+                if (eta > 0.0f) {
+                    float sigma_from_sq = sigma_from * sigma_from;
+                    float sigma_to_sq   = sigma_to * sigma_to;
+                    if (sigma_from_sq > 0.0f) {
+                        float term = sigma_to_sq * (sigma_from_sq - sigma_to_sq) / sigma_from_sq;
+                        if (term > 0.0f) {
+                            sigma_up = eta * std::sqrt(term);
+                        }
+                    }
+                    sigma_up            = std::min(sigma_up, sigma_to);
+                    float sigma_down_sq = sigma_to_sq - sigma_up * sigma_up;
+                    sigma_down          = sigma_down_sq > 0.0f ? std::sqrt(sigma_down_sq) : 0.0f;
+                }
+
+                if (sigma_down == 0.0f || !have_old_sigma) {
+                    float dt            = sigma_down - sigma_from;
+                    float* vec_x        = (float*)x->data;
+                    float* vec_denoised = (float*)denoised->data;
+
+                    for (int j = 0; j < ggml_nelements(x); j++) {
+                        float d  = (vec_x[j] - vec_denoised[j]) / sigma_from;
+                        vec_x[j] = vec_x[j] + d * dt;
+                    }
+                } else {
+                    float t      = t_fn(sigma_from);
+                    float t_old  = t_fn(old_sigma_down);
+                    float t_next = t_fn(sigma_down);
+                    float t_prev = t_fn(sigmas[i - 1]);
+                    float h      = t_next - t;
+                    float c2     = (t_prev - t_old) / h;
+
+                    float phi1_val = phi1_fn(-h);
+                    float phi2_val = phi2_fn(-h);
+                    float b1       = phi1_val - phi2_val / c2;
+                    float b2       = phi2_val / c2;
+
+                    if (!std::isfinite(b1)) {
+                        b1 = 0.0f;
+                    }
+                    if (!std::isfinite(b2)) {
+                        b2 = 0.0f;
+                    }
+
+                    float sigma_h           = sigma_fn(h);
+                    float* vec_x            = (float*)x->data;
+                    float* vec_denoised     = (float*)denoised->data;
+                    float* vec_old_denoised = (float*)old_denoised->data;
+
+                    for (int j = 0; j < ggml_nelements(x); j++) {
+                        vec_x[j] = sigma_h * vec_x[j] + h * (b1 * vec_denoised[j] + b2 * vec_old_denoised[j]);
+                    }
+                }
+
+                if (sigmas[i + 1] > 0 && sigma_up > 0.0f) {
+                    ggml_ext_im_set_randn_f32(noise, rng);
+                    float* vec_x     = (float*)x->data;
+                    float* vec_noise = (float*)noise->data;
+
+                    for (int j = 0; j < ggml_nelements(x); j++) {
+                        vec_x[j] = vec_x[j] + vec_noise[j] * sigma_up;
+                    }
+                }
+
+                float* vec_old_denoised = (float*)old_denoised->data;
+                float* vec_denoised     = (float*)denoised->data;
+                for (int j = 0; j < ggml_nelements(x); j++) {
+                    vec_old_denoised[j] = vec_denoised[j];
+                }
+
+                old_sigma_down = sigma_down;
+                have_old_sigma = true;
+            }
+        } break;
+        case RES_2S_SAMPLE_METHOD:  // Res 2s sampler
+        {
+            struct ggml_tensor* noise = ggml_dup_tensor(work_ctx, x);
+            struct ggml_tensor* x0    = ggml_dup_tensor(work_ctx, x);
+            struct ggml_tensor* x2    = ggml_dup_tensor(work_ctx, x);
+
+            const float c2 = 0.5f;
+            auto t_fn      = [](float sigma) -> float { return -logf(sigma); };
+            auto phi1_fn   = [](float t) -> float {
+                if (fabsf(t) < 1e-6f) {
+                    return 1.0f + t * 0.5f + (t * t) / 6.0f;
+                }
+                return (expf(t) - 1.0f) / t;
+            };
+            auto phi2_fn = [&](float t) -> float {
+                if (fabsf(t) < 1e-6f) {
+                    return 0.5f + t / 6.0f + (t * t) / 24.0f;
+                }
+                float phi1_val = phi1_fn(t);
+                return (phi1_val - 1.0f) / t;
+            };
+
+            for (int i = 0; i < steps; i++) {
+                float sigma_from = sigmas[i];
+                float sigma_to   = sigmas[i + 1];
+
+                ggml_tensor* denoised = model(x, sigma_from, -(i + 1));
+                if (denoised == nullptr) {
+                    return false;
+                }
+
+                float sigma_up   = 0.0f;
+                float sigma_down = sigma_to;
+                if (eta > 0.0f) {
+                    float sigma_from_sq = sigma_from * sigma_from;
+                    float sigma_to_sq   = sigma_to * sigma_to;
+                    if (sigma_from_sq > 0.0f) {
+                        float term = sigma_to_sq * (sigma_from_sq - sigma_to_sq) / sigma_from_sq;
+                        if (term > 0.0f) {
+                            sigma_up = eta * std::sqrt(term);
+                        }
+                    }
+                    sigma_up            = std::min(sigma_up, sigma_to);
+                    float sigma_down_sq = sigma_to_sq - sigma_up * sigma_up;
+                    sigma_down          = sigma_down_sq > 0.0f ? std::sqrt(sigma_down_sq) : 0.0f;
+                }
+
+                float* vec_x  = (float*)x->data;
+                float* vec_x0 = (float*)x0->data;
+                for (int j = 0; j < ggml_nelements(x); j++) {
+                    vec_x0[j] = vec_x[j];
+                }
+
+                if (sigma_down == 0.0f || sigma_from == 0.0f) {
+                    float* vec_denoised = (float*)denoised->data;
+                    for (int j = 0; j < ggml_nelements(x); j++) {
+                        vec_x[j] = vec_denoised[j];
+                    }
+                } else {
+                    float t      = t_fn(sigma_from);
+                    float t_next = t_fn(sigma_down);
+                    float h      = t_next - t;
+
+                    float a21      = c2 * phi1_fn(-h * c2);
+                    float phi1_val = phi1_fn(-h);
+                    float phi2_val = phi2_fn(-h);
+                    float b2       = phi2_val / c2;
+                    float b1       = phi1_val - b2;
+
+                    float sigma_c2 = expf(-(t + h * c2));
+
+                    float* vec_denoised = (float*)denoised->data;
+                    float* vec_x2       = (float*)x2->data;
+                    for (int j = 0; j < ggml_nelements(x); j++) {
+                        float eps1 = vec_denoised[j] - vec_x0[j];
+                        vec_x2[j]  = vec_x0[j] + h * a21 * eps1;
+                    }
+
+                    ggml_tensor* denoised2 = model(x2, sigma_c2, i + 1);
+                    if (denoised2 == nullptr) {
+                        return false;
+                    }
+                    float* vec_denoised2 = (float*)denoised2->data;
+
+                    for (int j = 0; j < ggml_nelements(x); j++) {
+                        float eps1 = vec_denoised[j] - vec_x0[j];
+                        float eps2 = vec_denoised2[j] - vec_x0[j];
+                        vec_x[j]   = vec_x0[j] + h * (b1 * eps1 + b2 * eps2);
+                    }
+                }
+
+                if (sigmas[i + 1] > 0 && sigma_up > 0.0f) {
+                    ggml_ext_im_set_randn_f32(noise, rng);
+                    float* vec_x     = (float*)x->data;
+                    float* vec_noise = (float*)noise->data;
+
+                    for (int j = 0; j < ggml_nelements(x); j++) {
+                        vec_x[j] = vec_x[j] + vec_noise[j] * sigma_up;
+                    }
+                }
+            }
+        } break;
 
         default:
             LOG_ERROR("Attempting to sample with nonexisting sample method %i", method);

src/diffusion_model.hpp

@@ -1,6 +1,7 @@
 #ifndef __DIFFUSION_MODEL_H__
 #define __DIFFUSION_MODEL_H__
 
+#include "anima.hpp"
 #include "flux.hpp"
 #include "mmdit.hpp"
 #include "qwen_image.hpp"
@@ -38,7 +39,7 @@ struct DiffusionModel {
     virtual size_t get_params_buffer_size()                                             = 0;
     virtual void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter){};
     virtual int64_t get_adm_in_channels()                            = 0;
-    virtual void set_flash_attn_enabled(bool enabled)                = 0;
+    virtual void set_flash_attention_enabled(bool enabled)           = 0;
     virtual void set_circular_axes(bool circular_x, bool circular_y) = 0;
 };
 
@@ -84,7 +85,7 @@ struct UNetModel : public DiffusionModel {
         return unet.unet.adm_in_channels;
     }
 
-    void set_flash_attn_enabled(bool enabled) {
+    void set_flash_attention_enabled(bool enabled) {
         unet.set_flash_attention_enabled(enabled);
     }
 
@@ -149,7 +150,7 @@ struct MMDiTModel : public DiffusionModel {
         return 768 + 1280;
     }
 
-    void set_flash_attn_enabled(bool enabled) {
+    void set_flash_attention_enabled(bool enabled) {
         mmdit.set_flash_attention_enabled(enabled);
     }
 
@@ -215,7 +216,7 @@ struct FluxModel : public DiffusionModel {
         return 768;
     }
 
-    void set_flash_attn_enabled(bool enabled) {
+    void set_flash_attention_enabled(bool enabled) {
         flux.set_flash_attention_enabled(enabled);
     }
 
@@ -242,6 +243,72 @@ struct FluxModel : public DiffusionModel {
     }
 };
 
+struct AnimaModel : public DiffusionModel {
+    std::string prefix;
+    Anima::AnimaRunner anima;
+
+    AnimaModel(ggml_backend_t backend,
+               bool offload_params_to_cpu,
+               const String2TensorStorage& tensor_storage_map = {},
+               const std::string prefix                       = "model.diffusion_model")
+        : prefix(prefix), anima(backend, offload_params_to_cpu, tensor_storage_map, prefix) {
+    }
+
+    std::string get_desc() override {
+        return anima.get_desc();
+    }
+
+    void alloc_params_buffer() override {
+        anima.alloc_params_buffer();
+    }
+
+    void free_params_buffer() override {
+        anima.free_params_buffer();
+    }
+
+    void free_compute_buffer() override {
+        anima.free_compute_buffer();
+    }
+
+    void get_param_tensors(std::map<std::string, struct ggml_tensor*>& tensors) override {
+        anima.get_param_tensors(tensors, prefix);
+    }
+
+    size_t get_params_buffer_size() override {
+        return anima.get_params_buffer_size();
+    }
+
+    void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) override {
+        anima.set_weight_adapter(adapter);
+    }
+
+    int64_t get_adm_in_channels() override {
+        return 768;
+    }
+
+    void set_flash_attention_enabled(bool enabled) {
+        anima.set_flash_attention_enabled(enabled);
+    }
+
+    void set_circular_axes(bool circular_x, bool circular_y) override {
+        anima.set_circular_axes(circular_x, circular_y);
+    }
+
+    bool compute(int n_threads,
+                 DiffusionParams diffusion_params,
+                 struct ggml_tensor** output     = nullptr,
+                 struct ggml_context* output_ctx = nullptr) override {
+        return anima.compute(n_threads,
+                             diffusion_params.x,
+                             diffusion_params.timesteps,
+                             diffusion_params.context,
+                             diffusion_params.c_concat,
+                             diffusion_params.y,
+                             output,
+                             output_ctx);
+    }
+};
+
 struct WanModel : public DiffusionModel {
     std::string prefix;
     WAN::WanRunner wan;
@@ -286,7 +353,7 @@ struct WanModel : public DiffusionModel {
         return 768;
     }
 
-    void set_flash_attn_enabled(bool enabled) {
+    void set_flash_attention_enabled(bool enabled) {
         wan.set_flash_attention_enabled(enabled);
     }
 
@@ -357,7 +424,7 @@ struct QwenImageModel : public DiffusionModel {
         return 768;
     }
 
-    void set_flash_attn_enabled(bool enabled) {
+    void set_flash_attention_enabled(bool enabled) {
         qwen_image.set_flash_attention_enabled(enabled);
     }
 
@@ -424,7 +491,7 @@ struct ZImageModel : public DiffusionModel {
         return 768;
     }
 
-    void set_flash_attn_enabled(bool enabled) {
+    void set_flash_attention_enabled(bool enabled) {
         z_image.set_flash_attention_enabled(enabled);
     }
 

src/esrgan.hpp

@@ -51,7 +51,7 @@ public:
         x_cat      = ggml_concat(ctx->ggml_ctx, x_cat, x4, 2);
         auto x5    = conv5->forward(ctx, x_cat);
 
-        x5 = ggml_add(ctx->ggml_ctx, ggml_scale(ctx->ggml_ctx, x5, 0.2f), x);
+        x5 = ggml_add(ctx->ggml_ctx, ggml_ext_scale(ctx->ggml_ctx, x5, 0.2f), x);
         return x5;
     }
 };
@@ -76,7 +76,7 @@ public:
         out      = rdb2->forward(ctx, out);
         out      = rdb3->forward(ctx, out);
 
-        out = ggml_add(ctx->ggml_ctx, ggml_scale(ctx->ggml_ctx, out, 0.2f), x);
+        out = ggml_add(ctx->ggml_ctx, ggml_ext_scale(ctx->ggml_ctx, out, 0.2f), x);
         return out;
     }
 };

src/flux.hpp

@@ -4,7 +4,7 @@
 #include <memory>
 #include <vector>
 
-#include "ggml_extend.hpp"
+#include "common_dit.hpp"
 #include "model.h"
 #include "rope.hpp"
 
@@ -103,11 +103,13 @@ namespace Flux {
             auto norm     = std::dynamic_pointer_cast<QKNorm>(blocks["norm"]);
 
             auto qkv         = qkv_proj->forward(ctx, x);
-            auto qkv_vec     = split_qkv(ctx->ggml_ctx, qkv);
+            int64_t head_dim = qkv->ne[0] / 3 / num_heads;
-            int64_t head_dim = qkv_vec[0]->ne[0] / num_heads;
+            auto q           = ggml_view_4d(ctx->ggml_ctx, qkv, head_dim, num_heads, qkv->ne[1], qkv->ne[2],
-            auto q           = ggml_reshape_4d(ctx->ggml_ctx, qkv_vec[0], head_dim, num_heads, qkv_vec[0]->ne[1], qkv_vec[0]->ne[2]);
+                                            qkv->nb[0] * head_dim, qkv->nb[1], qkv->nb[2], 0);
-            auto k           = ggml_reshape_4d(ctx->ggml_ctx, qkv_vec[1], head_dim, num_heads, qkv_vec[1]->ne[1], qkv_vec[1]->ne[2]);
+            auto k           = ggml_view_4d(ctx->ggml_ctx, qkv, head_dim, num_heads, qkv->ne[1], qkv->ne[2],
-            auto v           = ggml_reshape_4d(ctx->ggml_ctx, qkv_vec[2], head_dim, num_heads, qkv_vec[2]->ne[1], qkv_vec[2]->ne[2]);
+                                            qkv->nb[0] * head_dim, qkv->nb[1], qkv->nb[2], (qkv->nb[0]) * qkv->ne[0] / 3);
+            auto v           = ggml_view_4d(ctx->ggml_ctx, qkv, head_dim, num_heads, qkv->ne[1], qkv->ne[2],
+                                            qkv->nb[0] * head_dim, qkv->nb[1], qkv->nb[2], (qkv->nb[0]) * 2 * qkv->ne[0] / 3);
             q                = norm->query_norm(ctx, q);
             k                = norm->key_norm(ctx, k);
             return {q, k, v};
@@ -153,7 +155,7 @@ namespace Flux {
             if (use_mlp_silu_act) {
                 x = ggml_ext_silu_act(ctx->ggml_ctx, x);
             } else {
-                x = ggml_gelu_inplace(ctx->ggml_ctx, x);
+                x = ggml_ext_gelu(ctx->ggml_ctx, x, true);
             }
             x = mlp_2->forward(ctx, x);
             return x;
@@ -377,25 +379,22 @@ namespace Flux {
             auto v = ggml_concat(ctx->ggml_ctx, txt_v, img_v, 2);  // [N, n_txt_token + n_img_token, n_head, d_head]
 
             auto attn         = Rope::attention(ctx, q, k, v, pe, mask);  // [N, n_txt_token + n_img_token, n_head*d_head]
-            attn              = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, attn, 0, 2, 1, 3));  // [n_txt_token + n_img_token, N, hidden_size]
             auto txt_attn_out = ggml_view_3d(ctx->ggml_ctx,
                                              attn,
                                              attn->ne[0],
-                                             attn->ne[1],
                                              txt->ne[1],
+                                             attn->ne[2],
                                              attn->nb[1],
                                              attn->nb[2],
-                                             0);                                                                  // [n_txt_token, N, hidden_size]
+                                             0);  // [N, n_txt_token, hidden_size]
-            txt_attn_out      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, txt_attn_out, 0, 2, 1, 3));  // [N, n_txt_token, hidden_size]
             auto img_attn_out = ggml_view_3d(ctx->ggml_ctx,
                                              attn,
                                              attn->ne[0],
-                                             attn->ne[1],
                                              img->ne[1],
+                                             attn->ne[2],
                                              attn->nb[1],
                                              attn->nb[2],
-                                             attn->nb[2] * txt->ne[1]);                                           // [n_img_token, N, hidden_size]
+                                             txt->ne[1] * attn->nb[1]);  // [N, n_img_token, hidden_size]
-            img_attn_out      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, img_attn_out, 0, 2, 1, 3));  // [N, n_img_token, hidden_size]
 
             // calculate the img bloks
             img = ggml_add(ctx->ggml_ctx, img, ggml_mul(ctx->ggml_ctx, img_attn->post_attention(ctx, img_attn_out), img_mod1.gate));
@@ -492,43 +491,28 @@ namespace Flux {
             }
 
             auto x_mod   = Flux::modulate(ctx->ggml_ctx, pre_norm->forward(ctx, x), mod.shift, mod.scale);
-            auto qkv_mlp = linear1->forward(ctx, x_mod);                                                // [N, n_token, hidden_size * 3 + mlp_hidden_dim]
+            auto qkv_mlp = linear1->forward(ctx, x_mod);  // [N, n_token, hidden_size * 3 + mlp_hidden_dim*mlp_mult_factor]
-            qkv_mlp      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, qkv_mlp, 2, 0, 1, 3));  // [hidden_size * 3 + mlp_hidden_dim, N, n_token]
 
-            auto qkv = ggml_view_3d(ctx->ggml_ctx,
-                                    qkv_mlp,
-                                    qkv_mlp->ne[0],
-                                    qkv_mlp->ne[1],
-                                    hidden_size * 3,
-                                    qkv_mlp->nb[1],
-                                    qkv_mlp->nb[2],
-                                    0);                                                         // [hidden_size * 3 , N, n_token]
-            qkv      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, qkv, 1, 2, 0, 3));  // [N, n_token, hidden_size * 3]
-            auto mlp = ggml_view_3d(ctx->ggml_ctx,
-                                    qkv_mlp,
-                                    qkv_mlp->ne[0],
-                                    qkv_mlp->ne[1],
-                                    mlp_hidden_dim * mlp_mult_factor,
-                                    qkv_mlp->nb[1],
-                                    qkv_mlp->nb[2],
-                                    qkv_mlp->nb[2] * hidden_size * 3);                          // [mlp_hidden_dim*mlp_mult_factor , N, n_token]
-            mlp      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, mlp, 1, 2, 0, 3));  // [N, n_token, mlp_hidden_dim*mlp_mult_factor]
-
-            auto qkv_vec     = split_qkv(ctx->ggml_ctx, qkv);  // q,k,v: [N, n_token, hidden_size]
             int64_t head_dim = hidden_size / num_heads;
-            auto q           = ggml_reshape_4d(ctx->ggml_ctx, qkv_vec[0], head_dim, num_heads, qkv_vec[0]->ne[1], qkv_vec[0]->ne[2]);  // [N, n_token, n_head, d_head]
+
-            auto k           = ggml_reshape_4d(ctx->ggml_ctx, qkv_vec[1], head_dim, num_heads, qkv_vec[1]->ne[1], qkv_vec[1]->ne[2]);  // [N, n_token, n_head, d_head]
+            auto q = ggml_view_4d(ctx->ggml_ctx, qkv_mlp, head_dim, num_heads, qkv_mlp->ne[1], qkv_mlp->ne[2],
-            auto v           = ggml_reshape_4d(ctx->ggml_ctx, qkv_vec[2], head_dim, num_heads, qkv_vec[2]->ne[1], qkv_vec[2]->ne[2]);  // [N, n_token, n_head, d_head]
+                                  qkv_mlp->nb[0] * head_dim, qkv_mlp->nb[1], qkv_mlp->nb[2], 0);
+            auto k = ggml_view_4d(ctx->ggml_ctx, qkv_mlp, head_dim, num_heads, qkv_mlp->ne[1], qkv_mlp->ne[2],
+                                  qkv_mlp->nb[0] * head_dim, qkv_mlp->nb[1], qkv_mlp->nb[2], (qkv_mlp->nb[0]) * hidden_size);
+            auto v = ggml_view_4d(ctx->ggml_ctx, qkv_mlp, head_dim, num_heads, qkv_mlp->ne[1], qkv_mlp->ne[2],
+                                  qkv_mlp->nb[0] * head_dim, qkv_mlp->nb[1], qkv_mlp->nb[2], (qkv_mlp->nb[0]) * 2 * hidden_size);
+
             q         = norm->query_norm(ctx, q);
             k         = norm->key_norm(ctx, k);
             auto attn = Rope::attention(ctx, q, k, v, pe, mask);  // [N, n_token, hidden_size]
 
+            auto mlp = ggml_view_3d(ctx->ggml_ctx, qkv_mlp, mlp_hidden_dim * mlp_mult_factor, qkv_mlp->ne[1], qkv_mlp->ne[2], qkv_mlp->nb[1], qkv_mlp->nb[2], hidden_size * 3 * qkv_mlp->nb[0]);
             if (use_yak_mlp) {
                 mlp = ggml_ext_silu_act(ctx->ggml_ctx, mlp, false);
             } else if (use_mlp_silu_act) {
                 mlp = ggml_ext_silu_act(ctx->ggml_ctx, mlp);
             } else {
-                mlp = ggml_gelu_inplace(ctx->ggml_ctx, mlp);
+                mlp = ggml_ext_gelu(ctx->ggml_ctx, mlp, true);
             }
             auto attn_mlp = ggml_concat(ctx->ggml_ctx, attn, mlp, 0);  // [N, n_token, hidden_size + mlp_hidden_dim]
             auto output   = linear2->forward(ctx, attn_mlp);           // [N, n_token, hidden_size]
@@ -581,12 +565,9 @@ namespace Flux {
                 auto adaLN_modulation_1 = std::dynamic_pointer_cast<Linear>(blocks["adaLN_modulation.1"]);
 
                 auto m     = adaLN_modulation_1->forward(ctx, ggml_silu(ctx->ggml_ctx, c));  // [N, 2 * hidden_size]
-                m      = ggml_reshape_3d(ctx->ggml_ctx, m, c->ne[0], 2, c->ne[1]);              // [N, 2, hidden_size]
+                auto m_vec = ggml_ext_chunk(ctx->ggml_ctx, m, 2, 0);
-                m      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, m, 0, 2, 1, 3));  // [2, N, hidden_size]
+                shift      = m_vec[0];  // [N, hidden_size]
-
+                scale      = m_vec[1];  // [N, hidden_size]
-                int64_t offset = m->nb[1] * m->ne[1];
-                shift          = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 0);  // [N, hidden_size]
-                scale          = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 1);  // [N, hidden_size]
             }
 
             x = Flux::modulate(ctx->ggml_ctx, norm_final->forward(ctx, x), shift, scale);
@@ -748,7 +729,7 @@ namespace Flux {
         int nerf_depth           = 4;
         int nerf_max_freqs       = 8;
         bool use_x0              = false;
-        bool use_patch_size_32   = false;
+        bool fake_patch_size_x2  = false;
     };
 
     struct FluxParams {
@@ -786,7 +767,10 @@ namespace Flux {
         Flux(FluxParams params)
             : params(params) {
             if (params.version == VERSION_CHROMA_RADIANCE) {
-                std::pair<int, int> kernel_size = {16, 16};
+                std::pair<int, int> kernel_size = {params.patch_size, params.patch_size};
+                if (params.chroma_radiance_params.fake_patch_size_x2) {
+                    kernel_size = {params.patch_size / 2, params.patch_size / 2};
+                }
                 std::pair<int, int> stride = kernel_size;
 
                 blocks["img_in_patch"] = std::make_shared<Conv2d>(params.in_channels,
@@ -863,70 +847,6 @@ namespace Flux {
             }
         }
 
-        struct ggml_tensor* pad_to_patch_size(GGMLRunnerContext* ctx,
-                                              struct ggml_tensor* x) {
-            int64_t W = x->ne[0];
-            int64_t H = x->ne[1];
-
-            int pad_h = (params.patch_size - H % params.patch_size) % params.patch_size;
-            int pad_w = (params.patch_size - W % params.patch_size) % params.patch_size;
-            x         = ggml_ext_pad(ctx->ggml_ctx, x, pad_w, pad_h, 0, 0, ctx->circular_x_enabled, ctx->circular_y_enabled);
-            return x;
-        }
-
-        struct ggml_tensor* patchify(struct ggml_context* ctx,
-                                     struct ggml_tensor* x) {
-            // x: [N, C, H, W]
-            // return: [N, h*w, C * patch_size * patch_size]
-            int64_t N = x->ne[3];
-            int64_t C = x->ne[2];
-            int64_t H = x->ne[1];
-            int64_t W = x->ne[0];
-            int64_t p = params.patch_size;
-            int64_t h = H / params.patch_size;
-            int64_t w = W / params.patch_size;
-
-            GGML_ASSERT(h * p == H && w * p == W);
-
-            x = ggml_reshape_4d(ctx, x, p, w, p, h * C * N);       // [N*C*h, p, w, p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N*C*h, w, p, p]
-            x = ggml_reshape_4d(ctx, x, p * p, w * h, C, N);       // [N, C, h*w, p*p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N, h*w, C, p*p]
-            x = ggml_reshape_3d(ctx, x, p * p * C, w * h, N);      // [N, h*w, C*p*p]
-            return x;
-        }
-
-        struct ggml_tensor* process_img(GGMLRunnerContext* ctx,
-                                        struct ggml_tensor* x) {
-            // img = rearrange(x, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=patch_size, pw=patch_size)
-            x = pad_to_patch_size(ctx, x);
-            x = patchify(ctx->ggml_ctx, x);
-            return x;
-        }
-
-        struct ggml_tensor* unpatchify(struct ggml_context* ctx,
-                                       struct ggml_tensor* x,
-                                       int64_t h,
-                                       int64_t w) {
-            // x: [N, h*w, C*patch_size*patch_size]
-            // return: [N, C, H, W]
-            int64_t N = x->ne[2];
-            int64_t C = x->ne[0] / params.patch_size / params.patch_size;
-            int64_t H = h * params.patch_size;
-            int64_t W = w * params.patch_size;
-            int64_t p = params.patch_size;
-
-            GGML_ASSERT(C * p * p == x->ne[0]);
-
-            x = ggml_reshape_4d(ctx, x, p * p, C, w * h, N);       // [N, h*w, C, p*p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N, C, h*w, p*p]
-            x = ggml_reshape_4d(ctx, x, p, p, w, h * C * N);       // [N*C*h, w, p, p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N*C*h, p, w, p]
-            x = ggml_reshape_4d(ctx, x, W, H, C, N);               // [N, C, h*p, w*p]
-
-            return x;
-        }
-
         struct ggml_tensor* forward_orig(GGMLRunnerContext* ctx,
                                          struct ggml_tensor* img,
                                          struct ggml_tensor* txt,
@@ -1031,16 +951,14 @@ namespace Flux {
                 txt_img = block->forward(ctx, txt_img, vec, pe, txt_img_mask, ss_mods);
             }
 
-            txt_img = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, txt_img, 0, 2, 1, 3));  // [n_txt_token + n_img_token, N, hidden_size]
             img = ggml_view_3d(ctx->ggml_ctx,
                                txt_img,
                                txt_img->ne[0],
-                                   txt_img->ne[1],
                                img->ne[1],
+                               txt_img->ne[2],
                                txt_img->nb[1],
                                txt_img->nb[2],
-                                   txt_img->nb[2] * txt->ne[1]);                               // [n_img_token, N, hidden_size]
+                               txt->ne[1] * txt_img->nb[1]);  // [N, n_img_token, hidden_size]
-            img     = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, img, 0, 2, 1, 3));  // [N, n_img_token, hidden_size]
 
             if (final_layer) {
                 img = final_layer->forward(ctx, img, vec);  // (N, T, patch_size ** 2 * out_channels)
@@ -1079,10 +997,10 @@ namespace Flux {
             int pad_h      = (patch_size - H % patch_size) % patch_size;
             int pad_w      = (patch_size - W % patch_size) % patch_size;
 
-            auto img      = pad_to_patch_size(ctx, x);
+            auto img      = DiT::pad_to_patch_size(ctx, x, params.patch_size, params.patch_size);
             auto orig_img = img;
 
-            if (params.chroma_radiance_params.use_patch_size_32) {
+            if (params.chroma_radiance_params.fake_patch_size_x2) {
                 // It's supposed to be using GGML_SCALE_MODE_NEAREST, but this seems more stable
                 // Maybe the implementation of nearest-neighbor interpolation in ggml behaves differently than the one in PyTorch?
                 // img = F.interpolate(img, size=(H//2, W//2), mode="nearest")
@@ -1101,7 +1019,7 @@ namespace Flux {
             auto nerf_image_embedder   = std::dynamic_pointer_cast<NerfEmbedder>(blocks["nerf_image_embedder"]);
             auto nerf_final_layer_conv = std::dynamic_pointer_cast<NerfFinalLayerConv>(blocks["nerf_final_layer_conv"]);
 
-            auto nerf_pixels    = patchify(ctx->ggml_ctx, orig_img);  // [N, num_patches, C * patch_size * patch_size]
+            auto nerf_pixels    = DiT::patchify(ctx->ggml_ctx, orig_img, patch_size, patch_size);  // [N, num_patches, C * patch_size * patch_size]
             int64_t num_patches = nerf_pixels->ne[1];
             nerf_pixels         = ggml_reshape_3d(ctx->ggml_ctx,
                                                   nerf_pixels,
@@ -1121,7 +1039,7 @@ namespace Flux {
 
             img_dct = ggml_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, img_dct, 1, 0, 2, 3));                                 // [N*num_patches, nerf_hidden_size, patch_size*patch_size]
             img_dct = ggml_reshape_3d(ctx->ggml_ctx, img_dct, img_dct->ne[0] * img_dct->ne[1], num_patches, img_dct->ne[2] / num_patches);  // [N, num_patches, nerf_hidden_size*patch_size*patch_size]
-            img_dct = unpatchify(ctx->ggml_ctx, img_dct, (H + pad_h) / patch_size, (W + pad_w) / patch_size);                               // [N, nerf_hidden_size, H, W]
+            img_dct = DiT::unpatchify(ctx->ggml_ctx, img_dct, (H + pad_h) / patch_size, (W + pad_w) / patch_size, patch_size, patch_size);  // [N, nerf_hidden_size, H, W]
 
             out = nerf_final_layer_conv->forward(ctx, img_dct);  // [N, C, H, W]
 
@@ -1153,7 +1071,7 @@ namespace Flux {
             int pad_h      = (patch_size - H % patch_size) % patch_size;
             int pad_w      = (patch_size - W % patch_size) % patch_size;
 
-            auto img           = process_img(ctx, x);
+            auto img           = DiT::pad_and_patchify(ctx, x, patch_size, patch_size);
             int64_t img_tokens = img->ne[1];
 
             if (params.version == VERSION_FLUX_FILL) {
@@ -1161,8 +1079,8 @@ namespace Flux {
                 ggml_tensor* masked = ggml_view_4d(ctx->ggml_ctx, c_concat, c_concat->ne[0], c_concat->ne[1], C, 1, c_concat->nb[1], c_concat->nb[2], c_concat->nb[3], 0);
                 ggml_tensor* mask   = ggml_view_4d(ctx->ggml_ctx, c_concat, c_concat->ne[0], c_concat->ne[1], 8 * 8, 1, c_concat->nb[1], c_concat->nb[2], c_concat->nb[3], c_concat->nb[2] * C);
 
-                masked = process_img(ctx, masked);
+                masked = DiT::pad_and_patchify(ctx, masked, patch_size, patch_size);
-                mask   = process_img(ctx, mask);
+                mask   = DiT::pad_and_patchify(ctx, mask, patch_size, patch_size);
 
                 img = ggml_concat(ctx->ggml_ctx, img, ggml_concat(ctx->ggml_ctx, masked, mask, 0), 0);
             } else if (params.version == VERSION_FLEX_2) {
@@ -1171,21 +1089,21 @@ namespace Flux {
                 ggml_tensor* mask    = ggml_view_4d(ctx->ggml_ctx, c_concat, c_concat->ne[0], c_concat->ne[1], 1, 1, c_concat->nb[1], c_concat->nb[2], c_concat->nb[3], c_concat->nb[2] * C);
                 ggml_tensor* control = ggml_view_4d(ctx->ggml_ctx, c_concat, c_concat->ne[0], c_concat->ne[1], C, 1, c_concat->nb[1], c_concat->nb[2], c_concat->nb[3], c_concat->nb[2] * (C + 1));
 
-                masked  = process_img(ctx, masked);
+                masked  = DiT::pad_and_patchify(ctx, masked, patch_size, patch_size);
-                mask    = process_img(ctx, mask);
+                mask    = DiT::pad_and_patchify(ctx, mask, patch_size, patch_size);
-                control = process_img(ctx, control);
+                control = DiT::pad_and_patchify(ctx, control, patch_size, patch_size);
 
                 img = ggml_concat(ctx->ggml_ctx, img, ggml_concat(ctx->ggml_ctx, ggml_concat(ctx->ggml_ctx, masked, mask, 0), control, 0), 0);
             } else if (params.version == VERSION_FLUX_CONTROLS) {
                 GGML_ASSERT(c_concat != nullptr);
 
-                auto control = process_img(ctx, c_concat);
+                auto control = DiT::pad_and_patchify(ctx, c_concat, patch_size, patch_size);
                 img          = ggml_concat(ctx->ggml_ctx, img, control, 0);
             }
 
             if (ref_latents.size() > 0) {
                 for (ggml_tensor* ref : ref_latents) {
-                    ref = process_img(ctx, ref);
+                    ref = DiT::pad_and_patchify(ctx, ref, patch_size, patch_size);
                     img = ggml_concat(ctx->ggml_ctx, img, ref, 1);
                 }
             }
@@ -1193,13 +1111,11 @@ namespace Flux {
             auto out = forward_orig(ctx, img, context, timestep, y, guidance, pe, mod_index_arange, skip_layers);  // [N, num_tokens, C * patch_size * patch_size]
 
             if (out->ne[1] > img_tokens) {
-                out = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, out, 0, 2, 1, 3));  // [num_tokens, N, C * patch_size * patch_size]
+                out = ggml_view_3d(ctx->ggml_ctx, out, out->ne[0], img_tokens, out->ne[2], out->nb[1], out->nb[2], 0);
-                out = ggml_view_3d(ctx->ggml_ctx, out, out->ne[0], out->ne[1], img_tokens, out->nb[1], out->nb[2], 0);
+                out = ggml_cont(ctx->ggml_ctx, out);
-                out = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, out, 0, 2, 1, 3));  // [N, h*w, C * patch_size * patch_size]
             }
 
-            // rearrange(out, "b (h w) (c ph pw) -> b c (h ph) (w pw)", h=h_len, w=w_len, ph=2, pw=2)
+            out = DiT::unpatchify_and_crop(ctx->ggml_ctx, out, H, W, patch_size, patch_size);  // [N, C, H, W]
-            out = unpatchify(ctx->ggml_ctx, out, (H + pad_h) / patch_size, (W + pad_w) / patch_size);  // [N, C, H + pad_h, W + pad_w]
             return out;
         }
 
@@ -1304,6 +1220,7 @@ namespace Flux {
                 flux_params.use_mlp_silu_act = true;
             }
             int64_t head_dim                   = 0;
+            int64_t actual_radiance_patch_size = -1;
             for (auto pair : tensor_storage_map) {
                 std::string tensor_name = pair.first;
                 if (!starts_with(tensor_name, prefix))
@@ -1316,10 +1233,13 @@ namespace Flux {
                     flux_params.chroma_radiance_params.use_x0 = true;
                 }
                 if (tensor_name.find("__32x32__") != std::string::npos) {
-                    LOG_DEBUG("using patch size 32 prediction");
+                    LOG_DEBUG("using patch size 32");
-                    flux_params.chroma_radiance_params.use_patch_size_32 = true;
                     flux_params.patch_size = 32;
                 }
+                if (tensor_name.find("img_in_patch.weight") != std::string::npos) {
+                    actual_radiance_patch_size = pair.second.ne[0];
+                    LOG_DEBUG("actual radiance patch size: %d", actual_radiance_patch_size);
+                }
                 if (tensor_name.find("distilled_guidance_layer.in_proj.weight") != std::string::npos) {
                     // Chroma
                     flux_params.is_chroma = true;
@@ -1351,6 +1271,11 @@ namespace Flux {
                     head_dim = pair.second.ne[0];
                 }
             }
+            if (actual_radiance_patch_size > 0 && actual_radiance_patch_size != flux_params.patch_size) {
+                GGML_ASSERT(flux_params.patch_size == 2 * actual_radiance_patch_size);
+                LOG_DEBUG("using fake x2 patch size");
+                flux_params.chroma_radiance_params.fake_patch_size_x2 = true;
+            }
 
             flux_params.num_heads = static_cast<int>(flux_params.hidden_size / head_dim);
 

src/ggml_extend.hpp

@@ -491,12 +491,16 @@ __STATIC_INLINE__ void ggml_ext_tensor_split_2d(struct ggml_tensor* input,
     int64_t height   = output->ne[1];
     int64_t channels = output->ne[2];
     int64_t ne3      = output->ne[3];
+
+    int64_t input_width  = input->ne[0];
+    int64_t input_height = input->ne[1];
+
     GGML_ASSERT(input->type == GGML_TYPE_F32 && output->type == GGML_TYPE_F32);
     for (int iy = 0; iy < height; iy++) {
         for (int ix = 0; ix < width; ix++) {
             for (int k = 0; k < channels; k++) {
                 for (int l = 0; l < ne3; l++) {
-                    float value = ggml_ext_tensor_get_f32(input, ix + x, iy + y, k, l);
+                    float value = ggml_ext_tensor_get_f32(input, (ix + x) % input_width, (iy + y) % input_height, k, l);
                     ggml_ext_tensor_set_f32(output, value, ix, iy, k, l);
                 }
             }
@@ -516,6 +520,8 @@ __STATIC_INLINE__ void ggml_ext_tensor_merge_2d(struct ggml_tensor* input,
                                                 int y,
                                                 int overlap_x,
                                                 int overlap_y,
+                                                bool circular_x,
+                                                bool circular_y,
                                                 int x_skip = 0,
                                                 int y_skip = 0) {
     int64_t width    = input->ne[0];
@@ -533,12 +539,12 @@ __STATIC_INLINE__ void ggml_ext_tensor_merge_2d(struct ggml_tensor* input,
                 for (int l = 0; l < ne3; l++) {
                     float new_value = ggml_ext_tensor_get_f32(input, ix, iy, k, l);
                     if (overlap_x > 0 || overlap_y > 0) {  // blend colors in overlapped area
-                        float old_value = ggml_ext_tensor_get_f32(output, x + ix, y + iy, k, l);
+                        float old_value = ggml_ext_tensor_get_f32(output, (x + ix) % img_width, (y + iy) % img_height, k, l);
 
-                        const float x_f_0 = (overlap_x > 0 && x > 0) ? (ix - x_skip) / float(overlap_x) : 1;
+                        const float x_f_0 = (circular_x || (overlap_x > 0 && x > 0)) ? (ix - x_skip) / float(overlap_x) : 1;
-                        const float x_f_1 = (overlap_x > 0 && x < (img_width - width)) ? (width - ix) / float(overlap_x) : 1;
+                        const float x_f_1 = (circular_x || (overlap_x > 0 && x < (img_width - width))) ? (width - ix) / float(overlap_x) : 1;
-                        const float y_f_0 = (overlap_y > 0 && y > 0) ? (iy - y_skip) / float(overlap_y) : 1;
+                        const float y_f_0 = (circular_y || (overlap_y > 0 && y > 0)) ? (iy - y_skip) / float(overlap_y) : 1;
-                        const float y_f_1 = (overlap_y > 0 && y < (img_height - height)) ? (height - iy) / float(overlap_y) : 1;
+                        const float y_f_1 = (circular_y || (overlap_y > 0 && y < (img_height - height))) ? (height - iy) / float(overlap_y) : 1;
 
                         const float x_f = std::min(std::min(x_f_0, x_f_1), 1.f);
                         const float y_f = std::min(std::min(y_f_0, y_f_1), 1.f);
@@ -546,9 +552,9 @@ __STATIC_INLINE__ void ggml_ext_tensor_merge_2d(struct ggml_tensor* input,
                         ggml_ext_tensor_set_f32(
                             output,
                             old_value + new_value * smootherstep_f32(y_f) * smootherstep_f32(x_f),
-                            x + ix, y + iy, k, l);
+                            (x + ix) % img_width, (y + iy) % img_height, k, l);
                     } else {
-                        ggml_ext_tensor_set_f32(output, new_value, x + ix, y + iy, k, l);
+                        ggml_ext_tensor_set_f32(output, new_value, (x + ix) % img_width, (y + iy) % img_height, k, l);
                     }
                 }
             }
@@ -687,7 +693,8 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_slice(struct ggml_context* ctx,
                                                      struct ggml_tensor* x,
                                                      int dim,
                                                      int64_t start,
-                                                     int64_t end) {
+                                                     int64_t end,
+                                                     bool cont = true) {
     GGML_ASSERT(dim >= 0 && dim < 4);
     if (x->ne[dim] == 1) {
         return x;
@@ -702,27 +709,15 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_slice(struct ggml_context* ctx,
     GGML_ASSERT(start >= 0 && start < x->ne[dim]);
     GGML_ASSERT(end > start && end <= x->ne[dim]);
 
-    int perm[4] = {0, 1, 2, 3};
+    int64_t slice_size  = end - start;
-    for (int i = dim; i < 3; ++i)
+    int64_t slice_ne[4] = {x->ne[0], x->ne[1], x->ne[2], x->ne[3]};
-        perm[i] = perm[i + 1];
+    slice_ne[dim]       = slice_size;
-    perm[3] = dim;
 
-    int inv_perm[4];
+    x = ggml_view_4d(ctx, x,
-    for (int i = 0; i < 4; ++i)
+                     slice_ne[0], slice_ne[1], slice_ne[2], slice_ne[3],
-        inv_perm[perm[i]] = i;
+                     x->nb[1], x->nb[2], x->nb[3], start * x->nb[dim]);
 
-    if (dim != 3) {
+    if (cont) {
-        x = ggml_ext_torch_permute(ctx, x, perm[0], perm[1], perm[2], perm[3]);
-        x = ggml_cont(ctx, x);
-    }
-
-    x = ggml_view_4d(
-        ctx, x,
-        x->ne[0], x->ne[1], x->ne[2], end - start,
-        x->nb[1], x->nb[2], x->nb[3], x->nb[3] * start);
-
-    if (dim != 3) {
-        x = ggml_ext_torch_permute(ctx, x, inv_perm[0], inv_perm[1], inv_perm[2], inv_perm[3]);
         x = ggml_cont(ctx, x);
     }
 
@@ -778,16 +773,37 @@ __STATIC_INLINE__ ggml_tensor* ggml_ext_silu_act(ggml_context* ctx, ggml_tensor*
     return x;
 }
 
-typedef std::function<void(ggml_tensor*, ggml_tensor*, bool)> on_tile_process;
+typedef std::function<bool(ggml_tensor*, ggml_tensor*, bool)> on_tile_process;
 
 __STATIC_INLINE__ void sd_tiling_calc_tiles(int& num_tiles_dim,
                                             float& tile_overlap_factor_dim,
                                             int small_dim,
                                             int tile_size,
-                                            const float tile_overlap_factor) {
+                                            const float tile_overlap_factor,
+                                            bool circular) {
     int tile_overlap     = static_cast<int>(tile_size * tile_overlap_factor);
     int non_tile_overlap = tile_size - tile_overlap;
 
+    if (circular) {
+        // circular means the last and first tile are overlapping (wraping around)
+        num_tiles_dim = small_dim / non_tile_overlap;
+
+        if (num_tiles_dim < 1) {
+            num_tiles_dim = 1;
+        }
+
+        tile_overlap_factor_dim = (tile_size - small_dim / num_tiles_dim) / (float)tile_size;
+
+        // if single tile and tile_overlap_factor is not 0, add one to ensure we have at least two overlapping tiles
+        if (num_tiles_dim == 1 && tile_overlap_factor_dim > 0) {
+            num_tiles_dim++;
+            tile_overlap_factor_dim = 0.5;
+        }
+
+        return;
+    }
+    // else, non-circular means the last and first tile are not overlapping
+
     num_tiles_dim     = (small_dim - tile_overlap) / non_tile_overlap;
     int overshoot_dim = ((num_tiles_dim + 1) * non_tile_overlap + tile_overlap) % small_dim;
 
@@ -816,6 +832,8 @@ __STATIC_INLINE__ void sd_tiling_non_square(ggml_tensor* input,
                                             const int p_tile_size_x,
                                             const int p_tile_size_y,
                                             const float tile_overlap_factor,
+                                            const bool circular_x,
+                                            const bool circular_y,
                                             on_tile_process on_processing) {
     output = ggml_set_f32(output, 0);
 
@@ -840,11 +858,11 @@ __STATIC_INLINE__ void sd_tiling_non_square(ggml_tensor* input,
 
     int num_tiles_x;
     float tile_overlap_factor_x;
-    sd_tiling_calc_tiles(num_tiles_x, tile_overlap_factor_x, small_width, p_tile_size_x, tile_overlap_factor);
+    sd_tiling_calc_tiles(num_tiles_x, tile_overlap_factor_x, small_width, p_tile_size_x, tile_overlap_factor, circular_x);
 
     int num_tiles_y;
     float tile_overlap_factor_y;
-    sd_tiling_calc_tiles(num_tiles_y, tile_overlap_factor_y, small_height, p_tile_size_y, tile_overlap_factor);
+    sd_tiling_calc_tiles(num_tiles_y, tile_overlap_factor_y, small_height, p_tile_size_y, tile_overlap_factor, circular_y);
 
     LOG_DEBUG("num tiles : %d, %d ", num_tiles_x, num_tiles_y);
     LOG_DEBUG("optimal overlap : %f, %f (targeting %f)", tile_overlap_factor_x, tile_overlap_factor_y, tile_overlap_factor);
@@ -898,7 +916,7 @@ __STATIC_INLINE__ void sd_tiling_non_square(ggml_tensor* input,
     float last_time = 0.0f;
     for (int y = 0; y < small_height && !last_y; y += non_tile_overlap_y) {
         int dy = 0;
-        if (y + tile_size_y >= small_height) {
+        if (!circular_y && y + tile_size_y >= small_height) {
             int _y = y;
             y      = small_height - tile_size_y;
             dy     = _y - y;
@@ -909,7 +927,7 @@ __STATIC_INLINE__ void sd_tiling_non_square(ggml_tensor* input,
         }
         for (int x = 0; x < small_width && !last_x; x += non_tile_overlap_x) {
             int dx = 0;
-            if (x + tile_size_x >= small_width) {
+            if (!circular_x && x + tile_size_x >= small_width) {
                 int _x = x;
                 x      = small_width - tile_size_x;
                 dx     = _x - x;
@@ -929,12 +947,15 @@ __STATIC_INLINE__ void sd_tiling_non_square(ggml_tensor* input,
 
             int64_t t1 = ggml_time_ms();
             ggml_ext_tensor_split_2d(input, input_tile, x_in, y_in);
-            on_processing(input_tile, output_tile, false);
+            if (on_processing(input_tile, output_tile, false)) {
-            ggml_ext_tensor_merge_2d(output_tile, output, x_out, y_out, overlap_x_out, overlap_y_out, dx, dy);
+                ggml_ext_tensor_merge_2d(output_tile, output, x_out, y_out, overlap_x_out, overlap_y_out, circular_x, circular_y, dx, dy);
 
                 int64_t t2 = ggml_time_ms();
                 last_time  = (t2 - t1) / 1000.0f;
                 pretty_progress(tile_count, num_tiles, last_time);
+            } else {
+                LOG_ERROR("Failed to process patch %d at (%d, %d)", tile_count, x, y);
+            }
             tile_count++;
         }
         last_x = false;
@@ -950,8 +971,10 @@ __STATIC_INLINE__ void sd_tiling(ggml_tensor* input,
                                  const int scale,
                                  const int tile_size,
                                  const float tile_overlap_factor,
+                                 const bool circular_x,
+                                 const bool circular_y,
                                  on_tile_process on_processing) {
-    sd_tiling_non_square(input, output, scale, tile_size, tile_size, tile_overlap_factor, on_processing);
+    sd_tiling_non_square(input, output, scale, tile_size, tile_size, tile_overlap_factor, circular_x, circular_y, on_processing);
 }
 
 __STATIC_INLINE__ struct ggml_tensor* ggml_ext_group_norm_32(struct ggml_context* ctx,
@@ -960,6 +983,49 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_group_norm_32(struct ggml_context
     return ggml_group_norm(ctx, a, 32, eps);
 }
 
+__STATIC_INLINE__ struct ggml_tensor* ggml_ext_scale(struct ggml_context* ctx,
+                                                     struct ggml_tensor* x,
+                                                     float factor,
+                                                     bool inplace = false) {
+    if (!ggml_is_contiguous(x)) {
+        x = ggml_cont(ctx, x);
+    }
+    if (inplace) {
+        x = ggml_scale_inplace(ctx, x, factor);
+    } else {
+        x = ggml_scale(ctx, x, factor);
+    }
+    return x;
+}
+
+__STATIC_INLINE__ struct ggml_tensor* ggml_ext_gelu(struct ggml_context* ctx,
+                                                    struct ggml_tensor* x,
+                                                    bool inplace = false) {
+    if (!ggml_is_contiguous(x)) {
+        x = ggml_cont(ctx, x);
+    }
+    if (inplace) {
+        x = ggml_gelu_inplace(ctx, x);
+    } else {
+        x = ggml_gelu(ctx, x);
+    }
+    return x;
+}
+
+__STATIC_INLINE__ struct ggml_tensor* ggml_ext_gelu_quick(struct ggml_context* ctx,
+                                                          struct ggml_tensor* x,
+                                                          bool inplace = false) {
+    if (!ggml_is_contiguous(x)) {
+        x = ggml_cont(ctx, x);
+    }
+    if (inplace) {
+        x = ggml_gelu_quick_inplace(ctx, x);
+    } else {
+        x = ggml_gelu_quick(ctx, x);
+    }
+    return x;
+}
+
 __STATIC_INLINE__ struct ggml_tensor* ggml_ext_linear(struct ggml_context* ctx,
                                                       struct ggml_tensor* x,
                                                       struct ggml_tensor* w,
@@ -967,7 +1033,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_linear(struct ggml_context* ctx,
                                                       bool force_prec_f32 = false,
                                                       float scale         = 1.f) {
     if (scale != 1.f) {
-        x = ggml_scale(ctx, x, scale);
+        x = ggml_ext_scale(ctx, x, scale);
     }
     if (x->ne[2] * x->ne[3] > 1024) {
         // workaround: avoid ggml cuda error
@@ -986,7 +1052,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_linear(struct ggml_context* ctx,
         }
     }
     if (scale != 1.f) {
-        x = ggml_scale(ctx, x, 1.f / scale);
+        x = ggml_ext_scale(ctx, x, 1.f / scale);
     }
     if (b != nullptr) {
         x = ggml_add_inplace(ctx, x, b);
@@ -1055,7 +1121,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_conv_2d(struct ggml_context* ctx,
                                                        bool circular_y = false,
                                                        float scale     = 1.f) {
     if (scale != 1.f) {
-        x = ggml_scale(ctx, x, scale);
+        x = ggml_ext_scale(ctx, x, scale);
     }
     if (w->ne[2] != x->ne[2] && ggml_n_dims(w) == 2) {
         w = ggml_reshape_4d(ctx, w, 1, 1, w->ne[0], w->ne[1]);
@@ -1073,7 +1139,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_conv_2d(struct ggml_context* ctx,
         x = ggml_conv_2d(ctx, w, x, s0, s1, p0, p1, d0, d1);
     }
     if (scale != 1.f) {
-        x = ggml_scale(ctx, x, 1.f / scale);
+        x = ggml_ext_scale(ctx, x, 1.f / scale);
     }
     if (b != nullptr) {
         b = ggml_reshape_4d(ctx, b, 1, 1, b->ne[0], 1);
@@ -1171,7 +1237,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_full(struct ggml_context* ctx,
                                                     int64_t ne2,
                                                     int64_t ne3) {
     auto one = ggml_get_tensor(ctx, "ggml_runner_build_in_tensor:one");
-    auto t   = ggml_scale(ctx, one, value);                 // [1,]
+    auto t   = ggml_ext_scale(ctx, one, value);             // [1,]
     t        = ggml_repeat_4d(ctx, t, ne0, ne1, ne2, ne3);  // [ne0, ne1, ne2, ne3]
     return t;
 }
@@ -1184,6 +1250,11 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_zeros(struct ggml_context* ctx,
     return ggml_ext_full(ctx, 0.f, ne0, ne1, ne2, ne3);
 }
 
+__STATIC_INLINE__ struct ggml_tensor* ggml_ext_zeros_like(struct ggml_context* ctx,
+                                                          struct ggml_tensor* x) {
+    return ggml_ext_zeros(ctx, x->ne[0], x->ne[1], x->ne[2], x->ne[3]);
+}
+
 __STATIC_INLINE__ struct ggml_tensor* ggml_ext_ones(struct ggml_context* ctx,
                                                     int64_t ne0,
                                                     int64_t ne1,
@@ -1192,6 +1263,11 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_ones(struct ggml_context* ctx,
     return ggml_ext_full(ctx, 1.f, ne0, ne1, ne2, ne3);
 }
 
+__STATIC_INLINE__ struct ggml_tensor* ggml_ext_ones_like(struct ggml_context* ctx,
+                                                         struct ggml_tensor* x) {
+    return ggml_ext_ones(ctx, x->ne[0], x->ne[1], x->ne[2], x->ne[3]);
+}
+
 __STATIC_INLINE__ ggml_tensor* ggml_ext_cast_f32(ggml_context* ctx, ggml_tensor* a) {
 #ifdef SD_USE_VULKAN
     auto zero_index = ggml_get_tensor(ctx, "ggml_runner_build_in_tensor:zero_int");
@@ -1225,7 +1301,6 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_attention_ext(struct ggml_context
                                                              struct ggml_tensor* v,
                                                              int64_t n_head,
                                                              struct ggml_tensor* mask = nullptr,
-                                                             bool diag_mask_inf       = false,
                                                              bool skip_reshape        = false,
                                                              bool flash_attn          = false,
                                                              float kv_scale           = 1.0f) {  // avoid overflow
@@ -1271,7 +1346,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_attention_ext(struct ggml_context
             k_in = ggml_pad(ctx, k_in, 0, kv_pad, 0, 0);
         }
         if (kv_scale != 1.0f) {
-            k_in = ggml_scale(ctx, k_in, kv_scale);
+            k_in = ggml_ext_scale(ctx, k_in, kv_scale);
         }
         k_in = ggml_cast(ctx, k_in, GGML_TYPE_F16);
 
@@ -1281,7 +1356,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_attention_ext(struct ggml_context
             v_in = ggml_pad(ctx, v_in, 0, kv_pad, 0, 0);
         }
         if (kv_scale != 1.0f) {
-            v_in = ggml_scale(ctx, v_in, kv_scale);
+            v_in = ggml_ext_scale(ctx, v_in, kv_scale);
         }
         v_in = ggml_cast(ctx, v_in, GGML_TYPE_F16);
 
@@ -1313,7 +1388,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_attention_ext(struct ggml_context
         auto out = ggml_flash_attn_ext(ctx, q_in, k_in, v_in, mask_in, scale / kv_scale, 0, 0);
         ggml_flash_attn_ext_set_prec(out, GGML_PREC_F32);
         if (kv_scale != 1.0f) {
-            out = ggml_scale(ctx, out, 1.0f / kv_scale);
+            out = ggml_ext_scale(ctx, out, 1.0f / kv_scale);
         }
         return out;
     };
@@ -1353,9 +1428,6 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_attention_ext(struct ggml_context
         if (mask) {
             kq = ggml_add_inplace(ctx, kq, mask);
         }
-        if (diag_mask_inf) {
-            kq = ggml_diag_mask_inf_inplace(ctx, kq, 0);
-        }
         kq = ggml_soft_max_inplace(ctx, kq);
 
         kqv = ggml_mul_mat(ctx, v, kq);  // [N * n_head, L_q, d_head]
@@ -1523,7 +1595,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_timestep_embedding(
     int dim,
     int max_period    = 10000,
     float time_factor = 1.0f) {
-    timesteps = ggml_scale(ctx, timesteps, time_factor);
+    timesteps = ggml_ext_scale(ctx, timesteps, time_factor);
     return ggml_timestep_embedding(ctx, timesteps, dim, max_period);
 }
 
@@ -1549,7 +1621,7 @@ struct WeightAdapter {
             bool force_prec_f32 = false;
             float scale         = 1.f;
         } linear;
-        struct {
+        struct conv2d_params_t {
             int s0          = 1;
             int s1          = 1;
             int p0          = 0;
@@ -2572,7 +2644,7 @@ public:
     // x: [N, n_token, embed_dim]
     struct ggml_tensor* forward(GGMLRunnerContext* ctx,
                                 struct ggml_tensor* x,
-                                bool mask = false) {
+                                struct ggml_tensor* mask = nullptr) {
         auto out_proj = std::dynamic_pointer_cast<Linear>(blocks[out_proj_name]);
 
         ggml_tensor* q;
@@ -2595,11 +2667,180 @@ public:
             v = v_proj->forward(ctx, x);
         }
 
-        x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, n_head, nullptr, mask);  // [N, n_token, embed_dim]
+        x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, n_head, mask, false);  // [N, n_token, embed_dim]
 
         x = out_proj->forward(ctx, x);  // [N, n_token, embed_dim]
         return x;
     }
 };
 
+__STATIC_INLINE__ struct ggml_tensor* ggml_ext_lokr_forward(
+    struct ggml_context* ctx,
+    struct ggml_tensor* h,    // Input: [q, batch] or [W, H, q, batch]
+    struct ggml_tensor* w1,   // Outer C (Full rank)
+    struct ggml_tensor* w1a,  // Outer A (Low rank part 1)
+    struct ggml_tensor* w1b,  // Outer B (Low rank part 2)
+    struct ggml_tensor* w2,   // Inner BA (Full rank)
+    struct ggml_tensor* w2a,  // Inner A (Low rank part 1)
+    struct ggml_tensor* w2b,  // Inner B (Low rank part 2)
+    bool is_conv,
+    WeightAdapter::ForwardParams::conv2d_params_t conv_params,
+    float scale) {
+    GGML_ASSERT((w1 != NULL || (w1a != NULL && w1b != NULL)));
+    GGML_ASSERT((w2 != NULL || (w2a != NULL && w2b != NULL)));
+
+    int uq = (w1 != NULL) ? (int)w1->ne[0] : (int)w1a->ne[0];
+    int up = (w1 != NULL) ? (int)w1->ne[1] : (int)w1b->ne[1];
+
+    int q_actual = is_conv ? (int)h->ne[2] : (int)h->ne[0];
+    int vq       = q_actual / uq;
+
+    int vp = (w2 != NULL) ? (is_conv ? (int)w2->ne[3] : (int)w2->ne[1])
+                          : (int)w2a->ne[1];
+    GGML_ASSERT(q_actual == (uq * vq) && "Input dimension mismatch for LoKR split");
+
+    struct ggml_tensor* hb;
+
+    if (!is_conv) {
+        int batch          = (int)h->ne[1];
+        int merge_batch_uq = batch;
+        int merge_batch_vp = batch;
+
+#if SD_USE_VULKAN
+        if (batch > 1) {
+            // no access to backend here, worst case is slightly worse perfs for other backends when built alongside Vulkan backend
+            int max_batch    = 65535;
+            int max_batch_uq = max_batch / uq;
+            merge_batch_uq   = 1;
+            for (int i = max_batch_uq; i > 0; i--) {
+                if (batch % i == 0) {
+                    merge_batch_uq = i;
+                    break;
+                }
+            }
+
+            int max_batch_vp = max_batch / vp;
+            merge_batch_vp   = 1;
+            for (int i = max_batch_vp; i > 0; i--) {
+                if (batch % i == 0) {
+                    merge_batch_vp = i;
+                    break;
+                }
+            }
+        }
+#endif
+
+        struct ggml_tensor* h_split = ggml_reshape_3d(ctx, h, vq, uq * merge_batch_uq, batch / merge_batch_uq);
+        if (w2 != NULL) {
+            hb = ggml_mul_mat(ctx, w2, h_split);
+        } else {
+            hb = ggml_mul_mat(ctx, w2b, ggml_mul_mat(ctx, w2a, h_split));
+        }
+
+        if (batch > 1) {
+            hb = ggml_reshape_3d(ctx, hb, vp, uq, batch);
+        }
+        struct ggml_tensor* hb_t = ggml_cont(ctx, ggml_transpose(ctx, hb));
+        hb_t                     = ggml_reshape_3d(ctx, hb_t, uq, vp * merge_batch_vp, batch / merge_batch_vp);
+
+        struct ggml_tensor* hc_t;
+        if (w1 != NULL) {
+            hc_t = ggml_mul_mat(ctx, w1, hb_t);
+        } else {
+            hc_t = ggml_mul_mat(ctx, w1b, ggml_mul_mat(ctx, w1a, hb_t));
+        }
+
+        if (batch > 1) {
+            hc_t = ggml_reshape_3d(ctx, hc_t, up, vp, batch);
+        }
+
+        struct ggml_tensor* hc  = ggml_transpose(ctx, hc_t);
+        struct ggml_tensor* out = ggml_reshape_2d(ctx, ggml_cont(ctx, hc), up * vp, batch);
+        return ggml_scale(ctx, out, scale);
+    } else {
+        int batch = (int)h->ne[3];
+        // 1. Reshape input: [W, H, vq*uq, batch] -> [W, H, vq, uq * batch]
+        struct ggml_tensor* h_split = ggml_reshape_4d(ctx, h, h->ne[0], h->ne[1], vq, uq * batch);
+
+        if (w2 != NULL) {
+            hb = ggml_ext_conv_2d(ctx, h_split, w2, nullptr,
+                                  conv_params.s0,
+                                  conv_params.s1,
+                                  conv_params.p0,
+                                  conv_params.p1,
+                                  conv_params.d0,
+                                  conv_params.d1,
+                                  conv_params.direct,
+                                  conv_params.circular_x,
+                                  conv_params.circular_y,
+                                  conv_params.scale);
+        } else {
+            // swap a and b order for conv lora
+            struct ggml_tensor* a = w2b;
+            struct ggml_tensor* b = w2a;
+
+            // unpack conv2d weights if needed
+            if (ggml_n_dims(a) < 4) {
+                int k = (int)sqrt(a->ne[0] / h_split->ne[2]);
+                GGML_ASSERT(k * k * h_split->ne[2] == a->ne[0]);
+                a = ggml_reshape_4d(ctx, a, k, k, a->ne[0] / (k * k), a->ne[1]);
+            } else if (a->ne[2] != h_split->ne[2]) {
+                int k = (int)sqrt(a->ne[2] / h_split->ne[2]);
+                GGML_ASSERT(k * k * h_split->ne[2] == a->ne[2]);
+                a = ggml_reshape_4d(ctx, a, a->ne[0] * k, a->ne[1] * k, a->ne[2] / (k * k), a->ne[3]);
+            }
+            struct ggml_tensor* ha = ggml_ext_conv_2d(ctx, h_split, a, nullptr,
+                                                      conv_params.s0,
+                                                      conv_params.s1,
+                                                      conv_params.p0,
+                                                      conv_params.p1,
+                                                      conv_params.d0,
+                                                      conv_params.d1,
+                                                      conv_params.direct,
+                                                      conv_params.circular_x,
+                                                      conv_params.circular_y,
+                                                      conv_params.scale);
+
+            // not supporting lora_mid here
+            hb = ggml_ext_conv_2d(ctx,
+                                  ha,
+                                  b,
+                                  nullptr,
+                                  1,
+                                  1,
+                                  0,
+                                  0,
+                                  1,
+                                  1,
+                                  conv_params.direct,
+                                  conv_params.circular_x,
+                                  conv_params.circular_y,
+                                  conv_params.scale);
+        }
+
+        // Current hb shape: [W_out, H_out, vp, uq * batch]
+        int w_out = (int)hb->ne[0];
+        int h_out = (int)hb->ne[1];
+
+        // struct ggml_tensor* hb_cat = ggml_reshape_4d(ctx, hb, w_out , h_out , vp * uq, batch);
+        // [W_out, H_out, vp * uq,  batch]
+        // Now left to compute (W1 kr Id) * hb_cat == (W1 kr W2) cv h
+
+        // merge the uq groups of size vp*w_out*h_out
+        struct ggml_tensor* hb_merged = ggml_reshape_2d(ctx, hb, w_out * h_out * vp, uq * batch);
+        struct ggml_tensor* hc_t;
+        struct ggml_tensor* hb_merged_t = ggml_cont(ctx, ggml_transpose(ctx, hb_merged));
+        if (w1 != NULL) {
+            // Would be great to be able to transpose w1 instead to avoid transposing both hb and hc
+            hc_t = ggml_mul_mat(ctx, w1, hb_merged_t);
+        } else {
+            hc_t = ggml_mul_mat(ctx, w1b, ggml_mul_mat(ctx, w1a, hb_merged_t));
+        }
+        struct ggml_tensor* hc = ggml_transpose(ctx, hc_t);
+        // ungroup
+        struct ggml_tensor* out = ggml_reshape_4d(ctx, ggml_cont(ctx, hc), w_out, h_out, up * vp, batch);
+        return ggml_scale(ctx, out, scale);
+    }
+}
+
 #endif  // __GGML_EXTEND__HPP__

src/llm.hpp

@@ -19,6 +19,7 @@
 #include "json.hpp"
 #include "rope.hpp"
 #include "tokenize_util.h"
+#include "vocab/vocab.h"
 
 namespace LLM {
     constexpr int LLM_GRAPH_SIZE = 10240;
@@ -365,7 +366,7 @@ namespace LLM {
             if (merges_utf8_str.size() > 0) {
                 load_from_merges(merges_utf8_str);
             } else {
-                load_from_merges(ModelLoader::load_qwen2_merges());
+                load_from_merges(load_qwen2_merges());
             }
         }
     };
@@ -466,7 +467,7 @@ namespace LLM {
             if (merges_utf8_str.size() > 0 && vocab_utf8_str.size() > 0) {
                 load_from_merges(merges_utf8_str, vocab_utf8_str);
             } else {
-                load_from_merges(ModelLoader::load_mistral_merges(), ModelLoader::load_mistral_vocab_json());
+                load_from_merges(load_mistral_merges(), load_mistral_vocab_json());
             }
         }
     };
@@ -638,7 +639,7 @@ namespace LLM {
             x = ln_q->forward(ctx, x);
             x = ggml_reshape_2d(ctx->ggml_ctx, x, hidden_size, ggml_nelements(x) / hidden_size);
             x = mlp_0->forward(ctx, x);
-            x = ggml_gelu(ctx->ggml_ctx, x);
+            x = ggml_ext_gelu(ctx->ggml_ctx, x);
             x = mlp_2->forward(ctx, x);
             return x;
         }
@@ -881,7 +882,7 @@ namespace LLM {
             k = ggml_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, k, 0, 2, 1, 3));  // [N, num_kv_heads, n_token, head_dim]
             k = ggml_reshape_3d(ctx->ggml_ctx, k, k->ne[0], k->ne[1], k->ne[2] * k->ne[3]);      // [N*num_kv_heads, n_token, head_dim]
 
-            x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, num_heads, attention_mask, false, true, false);  // [N, n_token, hidden_size]
+            x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, num_heads, attention_mask, true, false);  // [N, n_token, hidden_size]
 
             x = out_proj->forward(ctx, x);  // [N, n_token, hidden_size]
             return x;

src/lora.hpp

@@ -195,7 +195,7 @@ struct LoraModel : public GGMLRunner {
             scale_value *= multiplier;
 
             auto curr_updown = ggml_ext_merge_lora(ctx, lora_down, lora_up, lora_mid);
-            curr_updown      = ggml_scale_inplace(ctx, curr_updown, scale_value);
+            curr_updown      = ggml_ext_scale(ctx, curr_updown, scale_value, true);
 
             if (updown == nullptr) {
                 updown = curr_updown;
@@ -235,7 +235,7 @@ struct LoraModel : public GGMLRunner {
             float scale_value = 1.0f;
             scale_value *= multiplier;
 
-            curr_updown = ggml_scale_inplace(ctx, curr_updown, scale_value);
+            curr_updown = ggml_ext_scale(ctx, curr_updown, scale_value, true);
 
             if (updown == nullptr) {
                 updown = curr_updown;
@@ -340,7 +340,7 @@ struct LoraModel : public GGMLRunner {
             struct ggml_tensor* updown_1 = ggml_ext_merge_lora(ctx, hada_1_down, hada_1_up, hada_1_mid);
             struct ggml_tensor* updown_2 = ggml_ext_merge_lora(ctx, hada_2_down, hada_2_up, hada_2_mid);
             auto curr_updown             = ggml_mul_inplace(ctx, updown_1, updown_2);
-            curr_updown                  = ggml_scale_inplace(ctx, curr_updown, scale_value);
+            curr_updown                  = ggml_ext_scale(ctx, curr_updown, scale_value, true);
             if (updown == nullptr) {
                 updown = curr_updown;
             } else {
@@ -456,7 +456,7 @@ struct LoraModel : public GGMLRunner {
             scale_value *= multiplier;
 
             auto curr_updown = ggml_ext_kronecker(ctx, lokr_w1, lokr_w2);
-            curr_updown      = ggml_scale_inplace(ctx, curr_updown, scale_value);
+            curr_updown      = ggml_ext_scale(ctx, curr_updown, scale_value, true);
 
             if (updown == nullptr) {
                 updown = curr_updown;
@@ -468,10 +468,10 @@ struct LoraModel : public GGMLRunner {
         return updown;
     }
 
-    ggml_tensor* get_weight_diff(const std::string& model_tensor_name, ggml_context* ctx, ggml_tensor* model_tensor, bool with_lora = true) {
+    ggml_tensor* get_weight_diff(const std::string& model_tensor_name, ggml_context* ctx, ggml_tensor* model_tensor, bool with_lora_and_lokr = true) {
         // lora
         ggml_tensor* diff = nullptr;
-        if (with_lora) {
+        if (with_lora_and_lokr) {
             diff = get_lora_weight_diff(model_tensor_name, ctx);
         }
         // diff
@@ -483,7 +483,7 @@ struct LoraModel : public GGMLRunner {
             diff = get_loha_weight_diff(model_tensor_name, ctx);
         }
         // lokr
-        if (diff == nullptr) {
+        if (diff == nullptr && with_lora_and_lokr) {
             diff = get_lokr_weight_diff(model_tensor_name, ctx);
         }
         if (diff != nullptr) {
@@ -514,6 +514,108 @@ struct LoraModel : public GGMLRunner {
             } else {
                 key = model_tensor_name + "." + std::to_string(index);
             }
+            bool is_conv2d = forward_params.op_type == WeightAdapter::ForwardParams::op_type_t::OP_CONV2D;
+
+            std::string lokr_w1_name   = "lora." + key + ".lokr_w1";
+            std::string lokr_w1_a_name = "lora." + key + ".lokr_w1_a";
+            // if either of these is found, then we have a lokr lora
+            auto iter   = lora_tensors.find(lokr_w1_name);
+            auto iter_a = lora_tensors.find(lokr_w1_a_name);
+            if (iter != lora_tensors.end() || iter_a != lora_tensors.end()) {
+                std::string lokr_w1_b_name = "lora." + key + ".lokr_w1_b";
+                std::string lokr_w2_name   = "lora." + key + ".lokr_w2";
+                std::string lokr_w2_a_name = "lora." + key + ".lokr_w2_a";
+                std::string lokr_w2_b_name = "lora." + key + ".lokr_w2_b";
+                std::string alpha_name     = "lora." + key + ".alpha";
+
+                ggml_tensor* lokr_w1   = nullptr;
+                ggml_tensor* lokr_w1_a = nullptr;
+                ggml_tensor* lokr_w1_b = nullptr;
+                ggml_tensor* lokr_w2   = nullptr;
+                ggml_tensor* lokr_w2_a = nullptr;
+                ggml_tensor* lokr_w2_b = nullptr;
+
+                if (iter != lora_tensors.end()) {
+                    lokr_w1 = iter->second;
+                }
+                iter = iter_a;
+                if (iter != lora_tensors.end()) {
+                    lokr_w1_a = iter->second;
+                }
+                iter = lora_tensors.find(lokr_w1_b_name);
+                if (iter != lora_tensors.end()) {
+                    lokr_w1_b = iter->second;
+                }
+
+                iter = lora_tensors.find(lokr_w2_name);
+                if (iter != lora_tensors.end()) {
+                    lokr_w2 = iter->second;
+                    if (is_conv2d && lokr_w2->type != GGML_TYPE_F16) {
+                        lokr_w2 = ggml_cast(ctx, lokr_w2, GGML_TYPE_F16);
+                    }
+                }
+                iter = lora_tensors.find(lokr_w2_a_name);
+                if (iter != lora_tensors.end()) {
+                    lokr_w2_a = iter->second;
+                    if (is_conv2d && lokr_w2_a->type != GGML_TYPE_F16) {
+                        lokr_w2_a = ggml_cast(ctx, lokr_w2_a, GGML_TYPE_F16);
+                    }
+                }
+                iter = lora_tensors.find(lokr_w2_b_name);
+                if (iter != lora_tensors.end()) {
+                    lokr_w2_b = iter->second;
+                    if (is_conv2d && lokr_w2_b->type != GGML_TYPE_F16) {
+                        lokr_w2_b = ggml_cast(ctx, lokr_w2_b, GGML_TYPE_F16);
+                    }
+                }
+
+                int rank = 1;
+                if (lokr_w1_b) {
+                    rank = (int)lokr_w1_b->ne[ggml_n_dims(lokr_w1_b) - 1];
+                }
+                if (lokr_w2_b) {
+                    rank = (int)lokr_w2_b->ne[ggml_n_dims(lokr_w2_b) - 1];
+                }
+
+                float scale_value = 1.0f;
+                iter              = lora_tensors.find(alpha_name);
+                if (iter != lora_tensors.end()) {
+                    float alpha = ggml_ext_backend_tensor_get_f32(iter->second);
+                    scale_value = alpha / rank;
+                    applied_lora_tensors.insert(alpha_name);
+                }
+
+                if (rank == 1) {
+                    scale_value = 1.0f;
+                }
+                scale_value *= multiplier;
+
+                auto curr_out_diff = ggml_ext_lokr_forward(ctx, x, lokr_w1, lokr_w1_a, lokr_w1_b, lokr_w2, lokr_w2_a, lokr_w2_b, is_conv2d, forward_params.conv2d, scale_value);
+                if (out_diff == nullptr) {
+                    out_diff = curr_out_diff;
+                } else {
+                    out_diff = ggml_concat(ctx, out_diff, curr_out_diff, 0);
+                }
+
+                if (lokr_w1)
+                    applied_lora_tensors.insert(lokr_w1_name);
+                if (lokr_w1_a)
+                    applied_lora_tensors.insert(lokr_w1_a_name);
+                if (lokr_w1_b)
+                    applied_lora_tensors.insert(lokr_w1_b_name);
+                if (lokr_w2)
+                    applied_lora_tensors.insert(lokr_w2_name);
+                if (lokr_w2_a)
+                    applied_lora_tensors.insert(lokr_w2_name);
+                if (lokr_w2_b)
+                    applied_lora_tensors.insert(lokr_w2_b_name);
+                applied_lora_tensors.insert(alpha_name);
+
+                index++;
+                continue;
+            }
+
+            // not a lokr, normal lora path
 
             std::string lora_down_name = "lora." + key + ".lora_down";
             std::string lora_up_name   = "lora." + key + ".lora_up";
@@ -525,9 +627,7 @@ struct LoraModel : public GGMLRunner {
             ggml_tensor* lora_mid  = nullptr;
             ggml_tensor* lora_down = nullptr;
 
-            bool is_conv2d = forward_params.op_type == WeightAdapter::ForwardParams::op_type_t::OP_CONV2D;
+            iter = lora_tensors.find(lora_up_name);
-
-            auto iter = lora_tensors.find(lora_up_name);
             if (iter != lora_tensors.end()) {
                 lora_up = iter->second;
                 if (is_conv2d && lora_up->type != GGML_TYPE_F16) {
@@ -634,7 +734,7 @@ struct LoraModel : public GGMLRunner {
                                       forward_params.conv2d.scale);
             }
 
-            auto curr_out_diff = ggml_scale_inplace(ctx, lx, scale_value);
+            auto curr_out_diff = ggml_ext_scale(ctx, lx, scale_value, true);
 
             if (out_diff == nullptr) {
                 out_diff = curr_out_diff;
@@ -741,9 +841,9 @@ public:
         : lora_models(lora_models) {
     }
 
-    ggml_tensor* patch_weight(ggml_context* ctx, ggml_tensor* weight, const std::string& weight_name, bool with_lora) {
+    ggml_tensor* patch_weight(ggml_context* ctx, ggml_tensor* weight, const std::string& weight_name, bool with_lora_and_lokr) {
         for (auto& lora_model : lora_models) {
-            ggml_tensor* diff = lora_model->get_weight_diff(weight_name, ctx, weight, with_lora);
+            ggml_tensor* diff = lora_model->get_weight_diff(weight_name, ctx, weight, with_lora_and_lokr);
             if (diff == nullptr) {
                 continue;
             }

src/ltxv.hpp

@@ -1,8 +1,7 @@
 #ifndef __LTXV_HPP__
 #define __LTXV_HPP__
 
-#include "common.hpp"
+#include "common_block.hpp"
-#include "ggml_extend.hpp"
 
 namespace LTXV {
 

src/mmdit.hpp

@@ -33,7 +33,7 @@ public:
         auto fc2 = std::dynamic_pointer_cast<Linear>(blocks["fc2"]);
 
         x = fc1->forward(ctx, x);
-        x = ggml_gelu_inplace(ctx->ggml_ctx, x);
+        x = ggml_ext_gelu(ctx->ggml_ctx, x, true);
         x = fc2->forward(ctx, x);
         return x;
     }
@@ -211,7 +211,7 @@ public:
     struct ggml_tensor* forward(GGMLRunnerContext* ctx,
                                 struct ggml_tensor* x) {
         auto qkv = pre_attention(ctx, x);
-        x        = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], num_heads, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
+        x        = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], num_heads, nullptr, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
         x        = post_attention(ctx, x);                                                                                                           // [N, n_token, dim]
         return x;
     }
@@ -284,23 +284,19 @@ public:
         auto attn2              = std::dynamic_pointer_cast<SelfAttention>(blocks["attn2"]);
         auto adaLN_modulation_1 = std::dynamic_pointer_cast<Linear>(blocks["adaLN_modulation.1"]);
 
-        int64_t n_mods = 9;
+        int n_mods = 9;
         auto m     = adaLN_modulation_1->forward(ctx, ggml_silu(ctx->ggml_ctx, c));  // [N, n_mods * hidden_size]
-        m              = ggml_reshape_3d(ctx->ggml_ctx, m, c->ne[0], n_mods, c->ne[1]);         // [N, n_mods, hidden_size]
+        auto m_vec = ggml_ext_chunk(ctx->ggml_ctx, m, n_mods, 0);
-        m              = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, m, 0, 2, 1, 3));  // [n_mods, N, hidden_size]
 
-        int64_t offset = m->nb[1] * m->ne[1];
+        auto shift_msa  = m_vec[0];  // [N, hidden_size]
-        auto shift_msa = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 0);  // [N, hidden_size]
+        auto scale_msa  = m_vec[1];  // [N, hidden_size]
-        auto scale_msa = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 1);  // [N, hidden_size]
+        auto gate_msa   = m_vec[2];  // [N, hidden_size]
-        auto gate_msa  = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 2);  // [N, hidden_size]
+        auto shift_mlp  = m_vec[3];  // [N, hidden_size]
-
+        auto scale_mlp  = m_vec[4];  // [N, hidden_size]
-        auto shift_mlp = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 3);  // [N, hidden_size]
+        auto gate_mlp   = m_vec[5];  // [N, hidden_size]
-        auto scale_mlp = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 4);  // [N, hidden_size]
+        auto shift_msa2 = m_vec[6];  // [N, hidden_size]
-        auto gate_mlp  = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 5);  // [N, hidden_size]
+        auto scale_msa2 = m_vec[7];  // [N, hidden_size]
-
+        auto gate_msa2  = m_vec[8];  // [N, hidden_size]
-        auto shift_msa2 = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 6);  // [N, hidden_size]
-        auto scale_msa2 = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 7);  // [N, hidden_size]
-        auto gate_msa2  = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 8);  // [N, hidden_size]
 
         auto x_norm = norm1->forward(ctx, x);
 
@@ -322,22 +318,20 @@ public:
         auto attn               = std::dynamic_pointer_cast<SelfAttention>(blocks["attn"]);
         auto adaLN_modulation_1 = std::dynamic_pointer_cast<Linear>(blocks["adaLN_modulation.1"]);
 
-        int64_t n_mods = 6;
+        int n_mods = 6;
         if (pre_only) {
             n_mods = 2;
         }
         auto m     = adaLN_modulation_1->forward(ctx, ggml_silu(ctx->ggml_ctx, c));  // [N, n_mods * hidden_size]
-        m      = ggml_reshape_3d(ctx->ggml_ctx, m, c->ne[0], n_mods, c->ne[1]);         // [N, n_mods, hidden_size]
+        auto m_vec = ggml_ext_chunk(ctx->ggml_ctx, m, n_mods, 0);
-        m      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, m, 0, 2, 1, 3));  // [n_mods, N, hidden_size]
 
-        int64_t offset = m->nb[1] * m->ne[1];
+        auto shift_msa = m_vec[0];  // [N, hidden_size]
-        auto shift_msa = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 0);  // [N, hidden_size]
+        auto scale_msa = m_vec[1];  // [N, hidden_size]
-        auto scale_msa = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 1);  // [N, hidden_size]
         if (!pre_only) {
-            auto gate_msa  = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 2);  // [N, hidden_size]
+            auto gate_msa  = m_vec[2];  // [N, hidden_size]
-            auto shift_mlp = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 3);  // [N, hidden_size]
+            auto shift_mlp = m_vec[3];  // [N, hidden_size]
-            auto scale_mlp = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 4);  // [N, hidden_size]
+            auto scale_mlp = m_vec[4];  // [N, hidden_size]
-            auto gate_mlp  = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 5);  // [N, hidden_size]
+            auto gate_mlp  = m_vec[5];  // [N, hidden_size]
 
             auto attn_in = modulate(ctx->ggml_ctx, norm1->forward(ctx, x), shift_msa, scale_msa);
 
@@ -439,8 +433,8 @@ public:
             auto qkv2          = std::get<1>(qkv_intermediates);
             auto intermediates = std::get<2>(qkv_intermediates);
 
-            auto attn_out  = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], num_heads, nullptr, false, false, ctx->flash_attn_enabled);     // [N, n_token, dim]
+            auto attn_out  = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], num_heads, nullptr, false, ctx->flash_attn_enabled);     // [N, n_token, dim]
-            auto attn2_out = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv2[0], qkv2[1], qkv2[2], num_heads, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
+            auto attn2_out = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv2[0], qkv2[1], qkv2[2], num_heads, nullptr, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
             x              = post_attention_x(ctx,
                                               attn_out,
                                               attn2_out,
@@ -456,7 +450,7 @@ public:
             auto qkv               = qkv_intermediates.first;
             auto intermediates     = qkv_intermediates.second;
 
-            auto attn_out = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], num_heads, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
+            auto attn_out = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], num_heads, nullptr, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
             x             = post_attention(ctx,
                                            attn_out,
                                            intermediates[0],
@@ -500,26 +494,24 @@ block_mixing(GGMLRunnerContext* ctx,
         qkv.push_back(ggml_concat(ctx->ggml_ctx, context_qkv[i], x_qkv[i], 1));
     }
 
-    auto attn         = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], x_block->num_heads, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_context + n_token, hidden_size]
+    auto attn = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, qkv[0], qkv[1], qkv[2], x_block->num_heads, nullptr, false, ctx->flash_attn_enabled);  // [N, n_context + n_token, hidden_size]
-    attn              = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, attn, 0, 2, 1, 3));                                                                          // [n_context + n_token, N, hidden_size]
+
     auto context_attn = ggml_view_3d(ctx->ggml_ctx,
                                      attn,
                                      attn->ne[0],
-                                     attn->ne[1],
                                      context->ne[1],
+                                     attn->ne[2],
                                      attn->nb[1],
                                      attn->nb[2],
-                                     0);                                                                  // [n_context, N, hidden_size]
+                                     0);  // [N, n_context, hidden_size]
-    context_attn      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, context_attn, 0, 2, 1, 3));  // [N, n_context, hidden_size]
     auto x_attn       = ggml_view_3d(ctx->ggml_ctx,
                                      attn,
                                      attn->ne[0],
-                                     attn->ne[1],
                                      x->ne[1],
+                                     attn->ne[2],
                                      attn->nb[1],
                                      attn->nb[2],
-                                     attn->nb[2] * context->ne[1]);                                 // [n_token, N, hidden_size]
+                                     context->ne[1] * attn->nb[1]);  // [N, n_token, hidden_size]
-    x_attn            = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, x_attn, 0, 2, 1, 3));  // [N, n_token, hidden_size]
 
     if (!context_block->pre_only) {
         context = context_block->post_attention(ctx,
@@ -534,7 +526,7 @@ block_mixing(GGMLRunnerContext* ctx,
     }
 
     if (x_block->self_attn) {
-        auto attn2 = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, x_qkv2[0], x_qkv2[1], x_qkv2[2], x_block->num_heads, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_token, hidden_size]
+        auto attn2 = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, x_qkv2[0], x_qkv2[1], x_qkv2[2], x_block->num_heads, nullptr, false, ctx->flash_attn_enabled);  // [N, n_token, hidden_size]
 
         x = x_block->post_attention_x(ctx,
                                       x_attn,
@@ -605,12 +597,9 @@ public:
         auto adaLN_modulation_1 = std::dynamic_pointer_cast<Linear>(blocks["adaLN_modulation.1"]);
 
         auto m     = adaLN_modulation_1->forward(ctx, ggml_silu(ctx->ggml_ctx, c));  // [N, 2 * hidden_size]
-        m      = ggml_reshape_3d(ctx->ggml_ctx, m, c->ne[0], 2, c->ne[1]);              // [N, 2, hidden_size]
+        auto m_vec = ggml_ext_chunk(ctx->ggml_ctx, m, 2, 0);
-        m      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, m, 0, 2, 1, 3));  // [2, N, hidden_size]
+        auto shift = m_vec[0];  // [N, hidden_size]
-
+        auto scale = m_vec[1];  // [N, hidden_size]
-        int64_t offset = m->nb[1] * m->ne[1];
-        auto shift     = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 0);  // [N, hidden_size]
-        auto scale     = ggml_view_2d(ctx->ggml_ctx, m, m->ne[0], m->ne[1], m->nb[1], offset * 1);  // [N, hidden_size]
 
         x = modulate(ctx->ggml_ctx, norm_final->forward(ctx, x), shift, scale);
         x = linear->forward(ctx, x);
@@ -756,28 +745,6 @@ public:
         return spatial_pos_embed;
     }
 
-    struct ggml_tensor* unpatchify(struct ggml_context* ctx,
-                                   struct ggml_tensor* x,
-                                   int64_t h,
-                                   int64_t w) {
-        // x: [N, H*W, patch_size * patch_size * C]
-        // return: [N, C, H, W]
-        int64_t n = x->ne[2];
-        int64_t c = out_channels;
-        int64_t p = patch_size;
-        h         = (h + 1) / p;
-        w         = (w + 1) / p;
-
-        GGML_ASSERT(h * w == x->ne[1]);
-
-        x = ggml_reshape_4d(ctx, x, c, p * p, w * h, n);       // [N, H*W, P*P, C]
-        x = ggml_cont(ctx, ggml_permute(ctx, x, 2, 0, 1, 3));  // [N, C, H*W, P*P]
-        x = ggml_reshape_4d(ctx, x, p, p, w, h * c * n);       // [N*C*H, W, P, P]
-        x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N*C*H, P, W, P]
-        x = ggml_reshape_4d(ctx, x, p * w, p * h, c, n);       // [N, C, H*P, W*P]
-        return x;
-    }
-
     struct ggml_tensor* forward_core_with_concat(GGMLRunnerContext* ctx,
                                                  struct ggml_tensor* x,
                                                  struct ggml_tensor* c_mod,
@@ -822,11 +789,11 @@ public:
         auto x_embedder = std::dynamic_pointer_cast<PatchEmbed>(blocks["x_embedder"]);
         auto t_embedder = std::dynamic_pointer_cast<TimestepEmbedder>(blocks["t_embedder"]);
 
-        int64_t w = x->ne[0];
+        int64_t W = x->ne[0];
-        int64_t h = x->ne[1];
+        int64_t H = x->ne[1];
 
         auto patch_embed = x_embedder->forward(ctx, x);                      // [N, H*W, hidden_size]
-        auto pos_embed   = cropped_pos_embed(ctx->ggml_ctx, h, w);           // [1, H*W, hidden_size]
+        auto pos_embed   = cropped_pos_embed(ctx->ggml_ctx, H, W);           // [1, H*W, hidden_size]
         x                = ggml_add(ctx->ggml_ctx, patch_embed, pos_embed);  // [N, H*W, hidden_size]
 
         auto c = t_embedder->forward(ctx, t);  // [N, hidden_size]
@@ -845,7 +812,7 @@ public:
 
         x = forward_core_with_concat(ctx, x, c, context, skip_layers);  // (N, H*W, patch_size ** 2 * out_channels)
 
-        x = unpatchify(ctx->ggml_ctx, x, h, w);  // [N, C, H, W]
+        x = DiT::unpatchify_and_crop(ctx->ggml_ctx, x, H, W, patch_size, patch_size, /*patch_last*/ false);  // [N, C, H, W]
 
         return x;
     }

src/model.cpp

@@ -16,10 +16,6 @@
 #include "model.h"
 #include "stable-diffusion.h"
 #include "util.h"
-#include "vocab.hpp"
-#include "vocab_mistral.hpp"
-#include "vocab_qwen.hpp"
-#include "vocab_umt5.hpp"
 
 #include "ggml-alloc.h"
 #include "ggml-backend.h"
@@ -376,7 +372,11 @@ bool ModelLoader::init_from_file(const std::string& file_path, const std::string
         LOG_INFO("load %s using checkpoint format", file_path.c_str());
         return init_from_ckpt_file(file_path, prefix);
     } else {
+        if (file_exists(file_path)) {
             LOG_WARN("unknown format %s", file_path.c_str());
+        } else {
+            LOG_WARN("file %s not found", file_path.c_str());
+        }
         return false;
     }
 }
@@ -1040,6 +1040,7 @@ SDVersion ModelLoader::get_sd_version() {
     int64_t patch_embedding_channels = 0;
     bool has_img_emb                 = false;
     bool has_middle_block_1          = false;
+    bool has_output_block_311        = false;
     bool has_output_block_71         = false;
 
     for (auto& [name, tensor_storage] : tensor_storage_map) {
@@ -1056,6 +1057,9 @@ SDVersion ModelLoader::get_sd_version() {
             if (tensor_storage.name.find("model.diffusion_model.transformer_blocks.0.img_mod.1.weight") != std::string::npos) {
                 return VERSION_QWEN_IMAGE;
             }
+            if (tensor_storage.name.find("llm_adapter.blocks.0.cross_attn.q_proj.weight") != std::string::npos) {
+                return VERSION_ANIMA;
+            }
             if (tensor_storage.name.find("model.diffusion_model.double_stream_modulation_img.lin.weight") != std::string::npos) {
                 is_flux2 = true;
             }
@@ -1100,6 +1104,9 @@ SDVersion ModelLoader::get_sd_version() {
             tensor_storage.name.find("unet.mid_block.resnets.1.") != std::string::npos) {
             has_middle_block_1 = true;
         }
+        if (tensor_storage.name.find("model.diffusion_model.output_blocks.3.1.transformer_blocks.1") != std::string::npos) {
+            has_output_block_311 = true;
+        }
         if (tensor_storage.name.find("model.diffusion_model.output_blocks.7.1") != std::string::npos) {
             has_output_block_71 = true;
         }
@@ -1138,6 +1145,9 @@ SDVersion ModelLoader::get_sd_version() {
             return VERSION_SDXL_PIX2PIX;
         }
         if (!has_middle_block_1) {
+            if (!has_output_block_311) {
+                return VERSION_SDXL_VEGA;
+            }
             return VERSION_SDXL_SSD1B;
         }
         return VERSION_SDXL;
@@ -1329,36 +1339,6 @@ void ModelLoader::set_wtype_override(ggml_type wtype, std::string tensor_type_ru
     }
 }
 
-std::string ModelLoader::load_merges() {
-    std::string merges_utf8_str(reinterpret_cast<const char*>(merges_utf8_c_str), sizeof(merges_utf8_c_str));
-    return merges_utf8_str;
-}
-
-std::string ModelLoader::load_qwen2_merges() {
-    std::string merges_utf8_str(reinterpret_cast<const char*>(qwen2_merges_utf8_c_str), sizeof(qwen2_merges_utf8_c_str));
-    return merges_utf8_str;
-}
-
-std::string ModelLoader::load_mistral_merges() {
-    std::string merges_utf8_str(reinterpret_cast<const char*>(mistral_merges_utf8_c_str), sizeof(mistral_merges_utf8_c_str));
-    return merges_utf8_str;
-}
-
-std::string ModelLoader::load_mistral_vocab_json() {
-    std::string json_str(reinterpret_cast<const char*>(mistral_vocab_json_utf8_c_str), sizeof(mistral_vocab_json_utf8_c_str));
-    return json_str;
-}
-
-std::string ModelLoader::load_t5_tokenizer_json() {
-    std::string json_str(reinterpret_cast<const char*>(t5_tokenizer_json_str), sizeof(t5_tokenizer_json_str));
-    return json_str;
-}
-
-std::string ModelLoader::load_umt5_tokenizer_json() {
-    std::string json_str(reinterpret_cast<const char*>(umt5_tokenizer_json_str), sizeof(umt5_tokenizer_json_str));
-    return json_str;
-}
-
 bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb, int n_threads_p, bool enable_mmap) {
     int64_t process_time_ms = 0;
     std::atomic<int64_t> read_time_ms(0);

src/model.h

@@ -32,6 +32,7 @@ enum SDVersion {
     VERSION_SDXL,
     VERSION_SDXL_INPAINT,
     VERSION_SDXL_PIX2PIX,
+    VERSION_SDXL_VEGA,
     VERSION_SDXL_SSD1B,
     VERSION_SVD,
     VERSION_SD3,
@@ -44,6 +45,7 @@ enum SDVersion {
     VERSION_WAN2_2_I2V,
     VERSION_WAN2_2_TI2V,
     VERSION_QWEN_IMAGE,
+    VERSION_ANIMA,
     VERSION_FLUX2,
     VERSION_FLUX2_KLEIN,
     VERSION_Z_IMAGE,
@@ -66,7 +68,7 @@ static inline bool sd_version_is_sd2(SDVersion version) {
 }
 
 static inline bool sd_version_is_sdxl(SDVersion version) {
-    if (version == VERSION_SDXL || version == VERSION_SDXL_INPAINT || version == VERSION_SDXL_PIX2PIX || version == VERSION_SDXL_SSD1B) {
+    if (version == VERSION_SDXL || version == VERSION_SDXL_INPAINT || version == VERSION_SDXL_PIX2PIX || version == VERSION_SDXL_SSD1B || version == VERSION_SDXL_VEGA) {
         return true;
     }
     return false;
@@ -121,6 +123,13 @@ static inline bool sd_version_is_qwen_image(SDVersion version) {
     return false;
 }
 
+static inline bool sd_version_is_anima(SDVersion version) {
+    if (version == VERSION_ANIMA) {
+        return true;
+    }
+    return false;
+}
+
 static inline bool sd_version_is_z_image(SDVersion version) {
     if (version == VERSION_Z_IMAGE) {
         return true;
@@ -145,6 +154,7 @@ static inline bool sd_version_is_dit(SDVersion version) {
         sd_version_is_sd3(version) ||
         sd_version_is_wan(version) ||
         sd_version_is_qwen_image(version) ||
+        sd_version_is_anima(version) ||
         sd_version_is_z_image(version)) {
         return true;
     }
@@ -330,13 +340,6 @@ public:
     bool tensor_should_be_converted(const TensorStorage& tensor_storage, ggml_type type);
     int64_t get_params_mem_size(ggml_backend_t backend, ggml_type type = GGML_TYPE_COUNT);
     ~ModelLoader() = default;
-
-    static std::string load_merges();
-    static std::string load_qwen2_merges();
-    static std::string load_mistral_merges();
-    static std::string load_mistral_vocab_json();
-    static std::string load_t5_tokenizer_json();
-    static std::string load_umt5_tokenizer_json();
 };
 
 #endif  // __MODEL_H__

src/name_conversion.cpp

@@ -653,6 +653,14 @@ std::string convert_diffusers_dit_to_original_lumina2(std::string name) {
     return name;
 }
 
+std::string convert_other_dit_to_original_anima(std::string name) {
+    static const std::string anima_net_prefix = "net.";
+    if (!starts_with(name, anima_net_prefix)) {
+        name = anima_net_prefix + name;
+    }
+    return name;
+}
+
 std::string convert_diffusion_model_name(std::string name, std::string prefix, SDVersion version) {
     if (sd_version_is_sd1(version) || sd_version_is_sd2(version)) {
         name = convert_diffusers_unet_to_original_sd1(name);
@@ -664,6 +672,8 @@ std::string convert_diffusion_model_name(std::string name, std::string prefix, S
         name = convert_diffusers_dit_to_original_flux(name);
     } else if (sd_version_is_z_image(version)) {
         name = convert_diffusers_dit_to_original_lumina2(name);
+    } else if (sd_version_is_anima(version)) {
+        name = convert_other_dit_to_original_anima(name);
     }
     return name;
 }
@@ -842,6 +852,7 @@ std::string convert_sep_to_dot(std::string name) {
         "conv_in",
         "conv_out",
         "lora_down",
+        "lora_mid",
         "lora_up",
         "diff_b",
         "hada_w1_a",
@@ -997,10 +1008,13 @@ std::string convert_tensor_name(std::string name, SDVersion version) {
     if (is_lora) {
         std::map<std::string, std::string> lora_suffix_map = {
             {".lora_down.weight", ".weight.lora_down"},
+            {".lora_mid.weight", ".weight.lora_mid"},
             {".lora_up.weight", ".weight.lora_up"},
             {".lora.down.weight", ".weight.lora_down"},
+            {".lora.mid.weight", ".weight.lora_mid"},
             {".lora.up.weight", ".weight.lora_up"},
             {"_lora.down.weight", ".weight.lora_down"},
+            {"_lora.mid.weight", ".weight.lora_mid"},
             {"_lora.up.weight", ".weight.lora_up"},
             {".lora_A.weight", ".weight.lora_down"},
             {".lora_B.weight", ".weight.lora_up"},

src/pmid.hpp

@@ -33,7 +33,7 @@ public:
         x = layer_norm->forward(ctx, x);
         // x = ggml_add(ctx, ggml_mul_mat(ctx, fc1_w, x),  fc1_b);
         x = fc1->forward(ctx, x);
-        x = ggml_gelu_inplace(ctx->ggml_ctx, x);
+        x = ggml_ext_gelu(ctx->ggml_ctx, x, true);
         x = fc2->forward(ctx, x);
         // x = ggml_add(ctx, ggml_mul_mat(ctx, fc2_w, x),  fc2_b);
         if (use_residue)
@@ -129,8 +129,8 @@ public:
         k             = reshape_tensor(ctx->ggml_ctx, k, heads);
         v             = reshape_tensor(ctx->ggml_ctx, v, heads);
         scale         = 1.f / sqrt(sqrt((float)dim_head));
-        k             = ggml_scale_inplace(ctx->ggml_ctx, k, scale);
+        k             = ggml_ext_scale(ctx->ggml_ctx, k, scale, true);
-        q             = ggml_scale_inplace(ctx->ggml_ctx, q, scale);
+        q             = ggml_ext_scale(ctx->ggml_ctx, q, scale, true);
         // auto weight = ggml_mul_mat(ctx, q, k);
         auto weight = ggml_mul_mat(ctx->ggml_ctx, k, q);  // NOTE order of mul is opposite to pytorch
 

src/qwen_image.hpp

@@ -3,9 +3,8 @@
 
 #include <memory>
 
-#include "common.hpp"
+#include "common_block.hpp"
 #include "flux.hpp"
-#include "ggml_extend.hpp"
 
 namespace Qwen {
     constexpr int QWEN_IMAGE_GRAPH_SIZE = 20480;
@@ -163,25 +162,24 @@ namespace Qwen {
             auto v = ggml_concat(ctx->ggml_ctx, txt_v, img_v, 2);  // [N, n_txt_token + n_img_token, n_head, d_head]
 
             auto attn         = Rope::attention(ctx, q, k, v, pe, mask, (1.0f / 128.f));  // [N, n_txt_token + n_img_token, n_head*d_head]
-            attn              = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, attn, 0, 2, 1, 3));  // [n_txt_token + n_img_token, N, hidden_size]
             auto txt_attn_out = ggml_view_3d(ctx->ggml_ctx,
                                              attn,
                                              attn->ne[0],
-                                             attn->ne[1],
                                              txt->ne[1],
+                                             attn->ne[2],
                                              attn->nb[1],
                                              attn->nb[2],
-                                             0);                                                                  // [n_txt_token, N, hidden_size]
+                                             0);  // [N, n_txt_token, n_head*d_head]
-            txt_attn_out      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, txt_attn_out, 0, 2, 1, 3));  // [N, n_txt_token, hidden_size]
             auto img_attn_out = ggml_view_3d(ctx->ggml_ctx,
                                              attn,
                                              attn->ne[0],
-                                             attn->ne[1],
                                              img->ne[1],
+                                             attn->ne[2],
                                              attn->nb[1],
                                              attn->nb[2],
-                                             attn->nb[2] * txt->ne[1]);                                           // [n_img_token, N, hidden_size]
+                                             txt->ne[1] * attn->nb[1]);  // [N, n_img_token, n_head*d_head]
-            img_attn_out      = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, img_attn_out, 0, 2, 1, 3));  // [N, n_img_token, hidden_size]
+            img_attn_out      = ggml_cont(ctx->ggml_ctx, img_attn_out);
+            txt_attn_out      = ggml_cont(ctx->ggml_ctx, txt_attn_out);
 
             img_attn_out = to_out_0->forward(ctx, img_attn_out);
             txt_attn_out = to_add_out->forward(ctx, txt_attn_out);
@@ -213,7 +211,7 @@ namespace Qwen {
 
             blocks["txt_norm1"] = std::shared_ptr<GGMLBlock>(new LayerNorm(dim, eps, false));
             blocks["txt_norm2"] = std::shared_ptr<GGMLBlock>(new LayerNorm(dim, eps, false));
-            blocks["txt_mlp"]   = std::shared_ptr<GGMLBlock>(new FeedForward(dim, dim, 4, FeedForward::Activation::GELU));
+            blocks["txt_mlp"]   = std::shared_ptr<GGMLBlock>(new FeedForward(dim, dim, 4, FeedForward::Activation::GELU, true));
 
             blocks["attn"] = std::shared_ptr<GGMLBlock>(new QwenImageAttention(dim,
                                                                                attention_head_dim,
@@ -391,69 +389,6 @@ namespace Qwen {
             blocks["proj_out"] = std::shared_ptr<GGMLBlock>(new Linear(inner_dim, params.patch_size * params.patch_size * params.out_channels));
         }
 
-        struct ggml_tensor* pad_to_patch_size(GGMLRunnerContext* ctx,
-                                              struct ggml_tensor* x) {
-            int64_t W = x->ne[0];
-            int64_t H = x->ne[1];
-
-            int pad_h = (params.patch_size - H % params.patch_size) % params.patch_size;
-            int pad_w = (params.patch_size - W % params.patch_size) % params.patch_size;
-            x         = ggml_ext_pad(ctx->ggml_ctx, x, pad_w, pad_h, 0, 0, ctx->circular_x_enabled, ctx->circular_y_enabled);
-            return x;
-        }
-
-        struct ggml_tensor* patchify(struct ggml_context* ctx,
-                                     struct ggml_tensor* x) {
-            // x: [N, C, H, W]
-            // return: [N, h*w, C * patch_size * patch_size]
-            int64_t N = x->ne[3];
-            int64_t C = x->ne[2];
-            int64_t H = x->ne[1];
-            int64_t W = x->ne[0];
-            int64_t p = params.patch_size;
-            int64_t h = H / params.patch_size;
-            int64_t w = W / params.patch_size;
-
-            GGML_ASSERT(h * p == H && w * p == W);
-
-            x = ggml_reshape_4d(ctx, x, p, w, p, h * C * N);       // [N*C*h, p, w, p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N*C*h, w, p, p]
-            x = ggml_reshape_4d(ctx, x, p * p, w * h, C, N);       // [N, C, h*w, p*p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N, h*w, C, p*p]
-            x = ggml_reshape_3d(ctx, x, p * p * C, w * h, N);      // [N, h*w, C*p*p]
-            return x;
-        }
-
-        struct ggml_tensor* process_img(GGMLRunnerContext* ctx,
-                                        struct ggml_tensor* x) {
-            x = pad_to_patch_size(ctx, x);
-            x = patchify(ctx->ggml_ctx, x);
-            return x;
-        }
-
-        struct ggml_tensor* unpatchify(struct ggml_context* ctx,
-                                       struct ggml_tensor* x,
-                                       int64_t h,
-                                       int64_t w) {
-            // x: [N, h*w, C*patch_size*patch_size]
-            // return: [N, C, H, W]
-            int64_t N = x->ne[2];
-            int64_t C = x->ne[0] / params.patch_size / params.patch_size;
-            int64_t H = h * params.patch_size;
-            int64_t W = w * params.patch_size;
-            int64_t p = params.patch_size;
-
-            GGML_ASSERT(C * p * p == x->ne[0]);
-
-            x = ggml_reshape_4d(ctx, x, p * p, C, w * h, N);       // [N, h*w, C, p*p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N, C, h*w, p*p]
-            x = ggml_reshape_4d(ctx, x, p, p, w, h * C * N);       // [N*C*h, w, p, p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));  // [N*C*h, p, w, p]
-            x = ggml_reshape_4d(ctx, x, W, H, C, N);               // [N, C, h*p, w*p]
-
-            return x;
-        }
-
         struct ggml_tensor* forward_orig(GGMLRunnerContext* ctx,
                                          struct ggml_tensor* x,
                                          struct ggml_tensor* timestep,
@@ -469,7 +404,7 @@ namespace Qwen {
 
             auto t_emb = time_text_embed->forward(ctx, timestep);
             if (params.zero_cond_t) {
-                auto t_emb_0 = time_text_embed->forward(ctx, ggml_ext_zeros(ctx->ggml_ctx, timestep->ne[0], timestep->ne[1], timestep->ne[2], timestep->ne[3]));
+                auto t_emb_0 = time_text_embed->forward(ctx, ggml_ext_zeros_like(ctx->ggml_ctx, timestep));
                 t_emb        = ggml_concat(ctx->ggml_ctx, t_emb, t_emb_0, 1);
             }
             auto img = img_in->forward(ctx, x);
@@ -513,19 +448,16 @@ namespace Qwen {
             int64_t C = x->ne[2];
             int64_t N = x->ne[3];
 
-            auto img           = process_img(ctx, x);
+            auto img           = DiT::pad_and_patchify(ctx, x, params.patch_size, params.patch_size);
             int64_t img_tokens = img->ne[1];
 
             if (ref_latents.size() > 0) {
                 for (ggml_tensor* ref : ref_latents) {
-                    ref = process_img(ctx, ref);
+                    ref = DiT::pad_and_patchify(ctx, ref, params.patch_size, params.patch_size);
                     img = ggml_concat(ctx->ggml_ctx, img, ref, 1);
                 }
             }
 
-            int64_t h_len = ((H + (params.patch_size / 2)) / params.patch_size);
-            int64_t w_len = ((W + (params.patch_size / 2)) / params.patch_size);
-
             auto out = forward_orig(ctx, img, timestep, context, pe, modulate_index);  // [N, h_len*w_len, ph*pw*C]
 
             if (out->ne[1] > img_tokens) {
@@ -534,11 +466,7 @@ namespace Qwen {
                 out = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, out, 0, 2, 1, 3));  // [N, h*w, C * patch_size * patch_size]
             }
 
-            out = unpatchify(ctx->ggml_ctx, out, h_len, w_len);  // [N, C, H + pad_h, W + pad_w]
+            out = DiT::unpatchify_and_crop(ctx->ggml_ctx, out, H, W, params.patch_size, params.patch_size);  // [N, C, H, W]
-
-            // slice
-            out = ggml_ext_slice(ctx->ggml_ctx, out, 1, 0, H);  // [N, C, H, W + pad_w]
-            out = ggml_ext_slice(ctx->ggml_ctx, out, 0, 0, W);  // [N, C, H, W]
 
             return out;
         }

src/rope.hpp

@@ -43,7 +43,7 @@ namespace Rope {
 
     __STATIC_INLINE__ std::vector<std::vector<float>> rope(const std::vector<float>& pos,
                                                            int dim,
-                                                           int theta,
+                                                           float theta,
                                                            const std::vector<int>& axis_wrap_dims = {}) {
         assert(dim % 2 == 0);
         int half_dim = dim / 2;
@@ -167,7 +167,7 @@ namespace Rope {
 
     __STATIC_INLINE__ std::vector<float> embed_nd(const std::vector<std::vector<float>>& ids,
                                                   int bs,
-                                                  int theta,
+                                                  const std::vector<float>& axis_thetas,
                                                   const std::vector<int>& axes_dim,
                                                   const std::vector<std::vector<int>>& wrap_dims = {}) {
         std::vector<std::vector<float>> trans_ids = transpose(ids);
@@ -188,8 +188,12 @@ namespace Rope {
             if (!wrap_dims.empty() && i < (int)wrap_dims.size()) {
                 axis_wrap_dims = wrap_dims[i];
             }
+            float axis_theta = 10000.0f;
+            if (!axis_thetas.empty()) {
+                axis_theta = axis_thetas[std::min(i, axis_thetas.size() - 1)];
+            }
             std::vector<std::vector<float>> rope_emb =
-                rope(trans_ids[i], axes_dim[i], theta, axis_wrap_dims);  // [bs*pos_len, axes_dim[i]/2 * 2 * 2]
+                rope(trans_ids[i], axes_dim[i], axis_theta, axis_wrap_dims);  // [bs*pos_len, axes_dim[i]/2 * 2 * 2]
             for (int b = 0; b < bs; ++b) {
                 for (int j = 0; j < pos_len; ++j) {
                     for (int k = 0; k < rope_emb[0].size(); ++k) {
@@ -203,6 +207,15 @@ namespace Rope {
         return flatten(emb);
     }
 
+    __STATIC_INLINE__ std::vector<float> embed_nd(const std::vector<std::vector<float>>& ids,
+                                                  int bs,
+                                                  float theta,
+                                                  const std::vector<int>& axes_dim,
+                                                  const std::vector<std::vector<int>>& wrap_dims = {}) {
+        std::vector<float> axis_thetas(axes_dim.size(), theta);
+        return embed_nd(ids, bs, axis_thetas, axes_dim, wrap_dims);
+    }
+
     __STATIC_INLINE__ std::vector<std::vector<float>> gen_refs_ids(int patch_size,
                                                                    int bs,
                                                                    int axes_dim_num,
@@ -332,7 +345,7 @@ namespace Rope {
                 }
             }
         }
-        return embed_nd(ids, bs, theta, axes_dim, wrap_dims);
+        return embed_nd(ids, bs, static_cast<float>(theta), axes_dim, wrap_dims);
     }
 
     __STATIC_INLINE__ std::vector<std::vector<float>> gen_qwen_image_ids(int h,
@@ -421,7 +434,7 @@ namespace Rope {
                 }
             }
         }
-        return embed_nd(ids, bs, theta, axes_dim, wrap_dims);
+        return embed_nd(ids, bs, static_cast<float>(theta), axes_dim, wrap_dims);
     }
 
     __STATIC_INLINE__ std::vector<std::vector<float>> gen_vid_ids(int t,
@@ -475,7 +488,7 @@ namespace Rope {
                                                     int theta,
                                                     const std::vector<int>& axes_dim) {
         std::vector<std::vector<float>> ids = gen_vid_ids(t, h, w, pt, ph, pw, bs);
-        return embed_nd(ids, bs, theta, axes_dim);
+        return embed_nd(ids, bs, static_cast<float>(theta), axes_dim);
     }
 
     __STATIC_INLINE__ std::vector<std::vector<float>> gen_qwen2vl_ids(int grid_h,
@@ -511,7 +524,7 @@ namespace Rope {
                                                         int theta,
                                                         const std::vector<int>& axes_dim) {
         std::vector<std::vector<float>> ids = gen_qwen2vl_ids(grid_h, grid_w, merge_size, window_index);
-        return embed_nd(ids, 1, theta, axes_dim);
+        return embed_nd(ids, 1, static_cast<float>(theta), axes_dim);
     }
 
     __STATIC_INLINE__ int bound_mod(int a, int m) {
@@ -584,7 +597,7 @@ namespace Rope {
             }
         }
 
-        return embed_nd(ids, bs, theta, axes_dim, wrap_dims);
+        return embed_nd(ids, bs, static_cast<float>(theta), axes_dim, wrap_dims);
     }
 
     __STATIC_INLINE__ struct ggml_tensor* apply_rope(struct ggml_context* ctx,
@@ -642,7 +655,7 @@ namespace Rope {
         q = apply_rope(ctx->ggml_ctx, q, pe, rope_interleaved);  // [N*n_head, L, d_head]
         k = apply_rope(ctx->ggml_ctx, k, pe, rope_interleaved);  // [N*n_head, L, d_head]
 
-        auto x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, v->ne[1], mask, false, true, ctx->flash_attn_enabled, kv_scale);  // [N, L, n_head*d_head]
+        auto x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, v->ne[1], mask, true, ctx->flash_attn_enabled, kv_scale);  // [N, L, n_head*d_head]
         return x;
     }
 };  // namespace Rope

src/spectrum.hpp

@@ -0,0 +1,195 @@
+#ifndef __SPECTRUM_HPP__
+#define __SPECTRUM_HPP__
+
+#include <cmath>
+#include <cstring>
+#include <vector>
+
+#include "ggml_extend.hpp"
+
+struct SpectrumConfig {
+    float w            = 0.40f;
+    int m              = 3;
+    float lam          = 1.0f;
+    int window_size    = 2;
+    float flex_window  = 0.50f;
+    int warmup_steps   = 4;
+    float stop_percent = 0.9f;
+};
+
+struct SpectrumState {
+    SpectrumConfig config;
+    int cnt                 = 0;
+    int num_cached          = 0;
+    float curr_ws           = 2.0f;
+    int K                   = 6;
+    int stop_step           = 0;
+    int total_steps_skipped = 0;
+
+    std::vector<std::vector<float>> H_buf;
+    std::vector<float> T_buf;
+
+    void init(const SpectrumConfig& cfg, size_t total_steps) {
+        config              = cfg;
+        cnt                 = 0;
+        num_cached          = 0;
+        curr_ws             = (float)cfg.window_size;
+        K                   = std::max(cfg.m + 1, 6);
+        stop_step           = (int)(cfg.stop_percent * (float)total_steps);
+        total_steps_skipped = 0;
+        H_buf.clear();
+        T_buf.clear();
+    }
+
+    float taus(int step_cnt) const {
+        return (step_cnt / 50.0f) * 2.0f - 1.0f;
+    }
+
+    bool should_predict() {
+        if (cnt < config.warmup_steps)
+            return false;
+        if (stop_step > 0 && cnt >= stop_step)
+            return false;
+        if ((int)H_buf.size() < 2)
+            return false;
+
+        int ws = std::max(1, (int)std::floor(curr_ws));
+        return (num_cached + 1) % ws != 0;
+    }
+
+    void update(const struct ggml_tensor* denoised) {
+        int64_t ne        = ggml_nelements(denoised);
+        const float* data = (const float*)denoised->data;
+
+        H_buf.emplace_back(data, data + ne);
+        T_buf.push_back(taus(cnt));
+
+        while ((int)H_buf.size() > K) {
+            H_buf.erase(H_buf.begin());
+            T_buf.erase(T_buf.begin());
+        }
+
+        if (cnt >= config.warmup_steps)
+            curr_ws += config.flex_window;
+
+        num_cached = 0;
+        cnt++;
+    }
+
+    void predict(struct ggml_tensor* denoised) {
+        int64_t F    = (int64_t)H_buf[0].size();
+        int K_curr   = (int)H_buf.size();
+        int M1       = config.m + 1;
+        float tau_at = taus(cnt);
+
+        // Design matrix X: K_curr x M1 (Chebyshev basis)
+        std::vector<float> X(K_curr * M1);
+        for (int i = 0; i < K_curr; i++) {
+            X[i * M1] = 1.0f;
+            if (M1 > 1)
+                X[i * M1 + 1] = T_buf[i];
+            for (int j = 2; j < M1; j++)
+                X[i * M1 + j] = 2.0f * T_buf[i] * X[i * M1 + j - 1] - X[i * M1 + j - 2];
+        }
+
+        // x_star: Chebyshev basis at current tau
+        std::vector<float> x_star(M1);
+        x_star[0] = 1.0f;
+        if (M1 > 1)
+            x_star[1] = tau_at;
+        for (int j = 2; j < M1; j++)
+            x_star[j] = 2.0f * tau_at * x_star[j - 1] - x_star[j - 2];
+
+        // XtX = X^T X + lambda I
+        std::vector<float> XtX(M1 * M1, 0.0f);
+        for (int i = 0; i < M1; i++) {
+            for (int j = 0; j < M1; j++) {
+                float sum = 0.0f;
+                for (int k = 0; k < K_curr; k++)
+                    sum += X[k * M1 + i] * X[k * M1 + j];
+                XtX[i * M1 + j] = sum + (i == j ? config.lam : 0.0f);
+            }
+        }
+
+        // Cholesky decomposition
+        std::vector<float> L(M1 * M1, 0.0f);
+        if (!cholesky_decompose(XtX.data(), L.data(), M1)) {
+            float trace = 0.0f;
+            for (int i = 0; i < M1; i++)
+                trace += XtX[i * M1 + i];
+            for (int i = 0; i < M1; i++)
+                XtX[i * M1 + i] += 1e-4f * trace / M1;
+            cholesky_decompose(XtX.data(), L.data(), M1);
+        }
+
+        // Solve XtX v = x_star
+        std::vector<float> v(M1);
+        cholesky_solve(L.data(), x_star.data(), v.data(), M1);
+
+        // Prediction weights per history entry
+        std::vector<float> weights(K_curr, 0.0f);
+        for (int k = 0; k < K_curr; k++)
+            for (int j = 0; j < M1; j++)
+                weights[k] += X[k * M1 + j] * v[j];
+
+        // Blend Chebyshev and Taylor predictions
+        float* out          = (float*)denoised->data;
+        float w_cheb        = config.w;
+        float w_taylor      = 1.0f - w_cheb;
+        const float* h_last = H_buf.back().data();
+        const float* h_prev = H_buf[H_buf.size() - 2].data();
+
+        for (int64_t f = 0; f < F; f++) {
+            float pred_cheb = 0.0f;
+            for (int k = 0; k < K_curr; k++)
+                pred_cheb += weights[k] * H_buf[k][f];
+
+            float pred_taylor = h_last[f] + 0.5f * (h_last[f] - h_prev[f]);
+
+            out[f] = w_taylor * pred_taylor + w_cheb * pred_cheb;
+        }
+
+        num_cached++;
+        total_steps_skipped++;
+        cnt++;
+    }
+
+private:
+    static bool cholesky_decompose(const float* A, float* L, int n) {
+        std::memset(L, 0, n * n * sizeof(float));
+        for (int i = 0; i < n; i++) {
+            for (int j = 0; j <= i; j++) {
+                float sum = 0.0f;
+                for (int k = 0; k < j; k++)
+                    sum += L[i * n + k] * L[j * n + k];
+                if (i == j) {
+                    float diag = A[i * n + i] - sum;
+                    if (diag <= 0.0f)
+                        return false;
+                    L[i * n + j] = std::sqrt(diag);
+                } else {
+                    L[i * n + j] = (A[i * n + j] - sum) / L[j * n + j];
+                }
+            }
+        }
+        return true;
+    }
+
+    static void cholesky_solve(const float* L, const float* b, float* x, int n) {
+        std::vector<float> y(n);
+        for (int i = 0; i < n; i++) {
+            float sum = 0.0f;
+            for (int j = 0; j < i; j++)
+                sum += L[i * n + j] * y[j];
+            y[i] = (b[i] - sum) / L[i * n + i];
+        }
+        for (int i = n - 1; i >= 0; i--) {
+            float sum = 0.0f;
+            for (int j = i + 1; j < n; j++)
+                sum += L[j * n + i] * x[j];
+            x[i] = (y[i] - sum) / L[i * n + i];
+        }
+    }
+};
+
+#endif  // __SPECTRUM_HPP__

src/stable-diffusion.cpp

@@ -16,6 +16,7 @@
 #include "esrgan.hpp"
 #include "lora.hpp"
 #include "pmid.hpp"
+#include "spectrum.hpp"
 #include "tae.hpp"
 #include "ucache.hpp"
 #include "vae.hpp"
@@ -35,6 +36,7 @@ const char* model_version_to_str[] = {
     "SDXL",
     "SDXL Inpaint",
     "SDXL Instruct-Pix2Pix",
+    "SDXL (Vega)",
     "SDXL (SSD1B)",
     "SVD",
     "SD3.x",
@@ -47,6 +49,7 @@ const char* model_version_to_str[] = {
     "Wan 2.2 I2V",
     "Wan 2.2 TI2V",
     "Qwen Image",
+    "Anima",
     "Flux.2",
     "Flux.2 klein",
     "Z-Image",
@@ -66,6 +69,8 @@ const char* sampling_methods_str[] = {
     "LCM",
     "DDIM \"trailing\"",
     "TCD",
+    "Res Multistep",
+    "Res 2s",
 };
 
 /*================================================== Helper Functions ================================================*/
@@ -93,6 +98,19 @@ void suppress_pp(int step, int steps, float time, void* data) {
     return;
 }
 
+static float get_cache_reuse_threshold(const sd_cache_params_t& params) {
+    float reuse_threshold = params.reuse_threshold;
+    if (reuse_threshold == INFINITY) {
+        if (params.mode == SD_CACHE_EASYCACHE) {
+            reuse_threshold = 0.2;
+        }
+        else if (params.mode == SD_CACHE_UCACHE) {
+            reuse_threshold = 1.0;
+        }
+    }
+    return std::max(0.0f, reuse_threshold);
+}
+
 /*=============================================== StableDiffusionGGML ================================================*/
 
 class StableDiffusionGGML {
@@ -104,13 +122,18 @@ public:
 
     SDVersion version;
     bool vae_decode_only         = false;
+    bool external_vae_is_invalid = false;
     bool free_params_immediately = false;
 
+    bool circular_x = false;
+    bool circular_y = false;
+
     std::shared_ptr<RNG> rng         = std::make_shared<PhiloxRNG>();
     std::shared_ptr<RNG> sampler_rng = nullptr;
     int n_threads                    = -1;
     float scale_factor               = 0.18215f;
     float shift_factor               = 0.f;
+    float default_flow_shift         = INFINITY;
 
     std::shared_ptr<Conditioner> cond_stage_model;
     std::shared_ptr<FrozenCLIPVisionEmbedder> clip_vision;  // for svd or wan2.1 i2v
@@ -318,6 +341,7 @@ public:
             LOG_INFO("loading vae from '%s'", sd_ctx_params->vae_path);
             if (!model_loader.init_from_file(sd_ctx_params->vae_path, "vae.")) {
                 LOG_WARN("loading vae from '%s' failed", sd_ctx_params->vae_path);
+                external_vae_is_invalid = true;
             }
         }
 
@@ -399,6 +423,7 @@ public:
             shift_factor = 0.1159f;
         } else if (sd_version_is_wan(version) ||
                    sd_version_is_qwen_image(version) ||
+                   sd_version_is_anima(version) ||
                    sd_version_is_flux2(version)) {
             scale_factor = 1.0f;
             shift_factor = 0.f;
@@ -442,7 +467,7 @@ public:
                     }
                 }
                 if (is_chroma) {
-                    if (sd_ctx_params->diffusion_flash_attn && sd_ctx_params->chroma_use_dit_mask) {
+                    if ((sd_ctx_params->flash_attn || sd_ctx_params->diffusion_flash_attn) && sd_ctx_params->chroma_use_dit_mask) {
                         LOG_WARN(
                             "!!!It looks like you are using Chroma with flash attention. "
                             "This is currently unsupported. "
@@ -529,6 +554,14 @@ public:
                                                                    "model.diffusion_model",
                                                                    version,
                                                                    sd_ctx_params->qwen_image_zero_cond_t);
+            } else if (sd_version_is_anima(version)) {
+                cond_stage_model = std::make_shared<AnimaConditioner>(clip_backend,
+                                                                      offload_params_to_cpu,
+                                                                      tensor_storage_map);
+                diffusion_model  = std::make_shared<AnimaModel>(backend,
+                                                               offload_params_to_cpu,
+                                                               tensor_storage_map,
+                                                               "model.diffusion_model");
             } else if (sd_version_is_z_image(version)) {
                 cond_stage_model = std::make_shared<LLMEmbedder>(clip_backend,
                                                                  offload_params_to_cpu,
@@ -568,14 +601,6 @@ public:
                 }
             }
 
-            if (sd_ctx_params->diffusion_flash_attn) {
-                LOG_INFO("Using flash attention in the diffusion model");
-                diffusion_model->set_flash_attn_enabled(true);
-                if (high_noise_diffusion_model) {
-                    high_noise_diffusion_model->set_flash_attn_enabled(true);
-                }
-            }
-
             cond_stage_model->alloc_params_buffer();
             cond_stage_model->get_param_tensors(tensors);
 
@@ -599,7 +624,7 @@ public:
             }
 
             if (!(use_tiny_autoencoder || version == VERSION_SDXS) || tae_preview_only) {
-                if (sd_version_is_wan(version) || sd_version_is_qwen_image(version)) {
+                if (sd_version_is_wan(version) || sd_version_is_qwen_image(version) || sd_version_is_anima(version)) {
                     first_stage_model = std::make_shared<WAN::WanVAERunner>(vae_backend,
                                                                             offload_params_to_cpu,
                                                                             tensor_storage_map,
@@ -623,11 +648,11 @@ public:
                         LOG_INFO("Using Conv2d direct in the vae model");
                         first_stage_model->set_conv2d_direct_enabled(true);
                     }
-                    if (version == VERSION_SDXL &&
+                    if (sd_version_is_sdxl(version) &&
-                        (strlen(SAFE_STR(sd_ctx_params->vae_path)) == 0 || sd_ctx_params->force_sdxl_vae_conv_scale)) {
+                        (strlen(SAFE_STR(sd_ctx_params->vae_path)) == 0 || sd_ctx_params->force_sdxl_vae_conv_scale || external_vae_is_invalid)) {
                         float vae_conv_2d_scale = 1.f / 32.f;
                         LOG_WARN(
-                            "No VAE specified with --vae or --force-sdxl-vae-conv-scale flag set, "
+                            "No valid VAE specified with --vae or --force-sdxl-vae-conv-scale flag set, "
                             "using Conv2D scale %.3f",
                             vae_conv_2d_scale);
                         first_stage_model->set_conv2d_scale(vae_conv_2d_scale);
@@ -637,7 +662,7 @@ public:
                 }
             }
             if (use_tiny_autoencoder || version == VERSION_SDXS) {
-                if (sd_version_is_wan(version) || sd_version_is_qwen_image(version)) {
+                if (sd_version_is_wan(version) || sd_version_is_qwen_image(version) || sd_version_is_anima(version)) {
                     tae_first_stage = std::make_shared<TinyVideoAutoEncoder>(vae_backend,
                                                                              offload_params_to_cpu,
                                                                              tensor_storage_map,
@@ -722,6 +747,28 @@ public:
                 pmid_model->get_param_tensors(tensors, "pmid");
             }
 
+            if (sd_ctx_params->flash_attn) {
+                LOG_INFO("Using flash attention");
+                cond_stage_model->set_flash_attention_enabled(true);
+                if (clip_vision) {
+                    clip_vision->set_flash_attention_enabled(true);
+                }
+                if (first_stage_model) {
+                    first_stage_model->set_flash_attention_enabled(true);
+                }
+                if (tae_first_stage) {
+                    tae_first_stage->set_flash_attention_enabled(true);
+                }
+            }
+
+            if (sd_ctx_params->flash_attn || sd_ctx_params->diffusion_flash_attn) {
+                LOG_INFO("Using flash attention in the diffusion model");
+                diffusion_model->set_flash_attention_enabled(true);
+                if (high_noise_diffusion_model) {
+                    high_noise_diffusion_model->set_flash_attention_enabled(true);
+                }
+            }
+
             diffusion_model->set_circular_axes(sd_ctx_params->circular_x, sd_ctx_params->circular_y);
             if (high_noise_diffusion_model) {
                 high_noise_diffusion_model->set_circular_axes(sd_ctx_params->circular_x, sd_ctx_params->circular_y);
@@ -729,12 +776,8 @@ public:
             if (control_net) {
                 control_net->set_circular_axes(sd_ctx_params->circular_x, sd_ctx_params->circular_y);
             }
-            if (first_stage_model) {
+            circular_x = sd_ctx_params->circular_x;
-                first_stage_model->set_circular_axes(sd_ctx_params->circular_x, sd_ctx_params->circular_y);
+            circular_y = sd_ctx_params->circular_y;
-            }
-            if (tae_first_stage) {
-                tae_first_stage->set_circular_axes(sd_ctx_params->circular_x, sd_ctx_params->circular_y);
-            }
         }
 
         struct ggml_init_params params;
@@ -862,7 +905,6 @@ public:
         // init denoiser
         {
             prediction_t pred_type = sd_ctx_params->prediction;
-            float flow_shift       = sd_ctx_params->flow_shift;
 
             if (pred_type == PREDICTION_COUNT) {
                 if (sd_version_is_sd2(version)) {
@@ -885,24 +927,22 @@ public:
                 } else if (sd_version_is_sd3(version) ||
                            sd_version_is_wan(version) ||
                            sd_version_is_qwen_image(version) ||
+                           sd_version_is_anima(version) ||
                            sd_version_is_z_image(version)) {
                     pred_type = FLOW_PRED;
-                    if (flow_shift == INFINITY) {
                     if (sd_version_is_wan(version)) {
-                            flow_shift = 5.f;
+                        default_flow_shift = 5.f;
                     } else {
-                            flow_shift = 3.f;
+                        default_flow_shift = 3.f;
-                        }
                     }
                 } else if (sd_version_is_flux(version)) {
                     pred_type = FLUX_FLOW_PRED;
 
-                    if (flow_shift == INFINITY) {
+                    default_flow_shift = 1.0f;  // TODO: validate
-                        flow_shift = 1.0f;  // TODO: validate
                     for (const auto& [name, tensor_storage] : tensor_storage_map) {
                         if (starts_with(name, "model.diffusion_model.guidance_in.in_layer.weight")) {
-                                flow_shift = 1.15f;
+                            default_flow_shift = 1.15f;
-                            }
+                            break;
                         }
                     }
                 } else if (sd_version_is_flux2(version)) {
@@ -926,12 +966,12 @@ public:
                     break;
                 case FLOW_PRED: {
                     LOG_INFO("running in FLOW mode");
-                    denoiser = std::make_shared<DiscreteFlowDenoiser>(flow_shift);
+                    denoiser = std::make_shared<DiscreteFlowDenoiser>();
                     break;
                 }
                 case FLUX_FLOW_PRED: {
                     LOG_INFO("running in Flux FLOW mode");
-                    denoiser = std::make_shared<FluxFlowDenoiser>(flow_shift);
+                    denoiser = std::make_shared<FluxFlowDenoiser>();
                     break;
                 }
                 case FLUX2_FLOW_PRED: {
@@ -1071,6 +1111,18 @@ public:
         cond_stage_lora_models.clear();
         diffusion_lora_models.clear();
         first_stage_lora_models.clear();
+        if (cond_stage_model) {
+            cond_stage_model->set_weight_adapter(nullptr);
+        }
+        if (diffusion_model) {
+            diffusion_model->set_weight_adapter(nullptr);
+        }
+        if (high_noise_diffusion_model) {
+            high_noise_diffusion_model->set_weight_adapter(nullptr);
+        }
+        if (first_stage_model) {
+            first_stage_model->set_weight_adapter(nullptr);
+        }
         if (lora_state.empty()) {
             return;
         }
@@ -1440,7 +1492,7 @@ public:
         sd_progress_cb_t cb = sd_get_progress_callback();
         void* cbd           = sd_get_progress_callback_data();
         sd_set_progress_callback((sd_progress_cb_t)suppress_pp, nullptr);
-        sd_tiling(input, output, scale, tile_size, tile_overlap_factor, on_processing);
+        sd_tiling(input, output, scale, tile_size, tile_overlap_factor, circular_x, circular_y, on_processing);
         sd_set_progress_callback(cb, cbd);
     }
 
@@ -1487,7 +1539,7 @@ public:
                 } else if (sd_version_is_flux(version) || sd_version_is_z_image(version)) {
                     latent_rgb_proj = flux_latent_rgb_proj;
                     latent_rgb_bias = flux_latent_rgb_bias;
-                } else if (sd_version_is_wan(version) || sd_version_is_qwen_image(version)) {
+                } else if (sd_version_is_wan(version) || sd_version_is_qwen_image(version) || sd_version_is_anima(version)) {
                     latent_rgb_proj = wan_21_latent_rgb_proj;
                     latent_rgb_bias = wan_21_latent_rgb_bias;
                 } else {
@@ -1541,7 +1593,7 @@ public:
                 if (vae_tiling_params.enabled) {
                     // split latent in 32x32 tiles and compute in several steps
                     auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
-                        first_stage_model->compute(n_threads, in, true, &out, nullptr);
+                        return first_stage_model->compute(n_threads, in, true, &out, nullptr);
                     };
                     silent_tiling(latents, result, get_vae_scale_factor(), 32, 0.5f, on_tiling);
 
@@ -1560,7 +1612,7 @@ public:
                 if (vae_tiling_params.enabled) {
                     // split latent in 64x64 tiles and compute in several steps
                     auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
-                        tae_first_stage->compute(n_threads, in, true, &out, nullptr);
+                        return tae_first_stage->compute(n_threads, in, true, &out, nullptr);
                     };
                     silent_tiling(latents, result, get_vae_scale_factor(), 64, 0.5f, on_tiling);
                 } else {
@@ -1649,9 +1701,11 @@ public:
         EasyCacheState easycache_state;
         UCacheState ucache_state;
         CacheDitConditionState cachedit_state;
+        SpectrumState spectrum_state;
         bool easycache_enabled = false;
         bool ucache_enabled    = false;
         bool cachedit_enabled  = false;
+        bool spectrum_enabled  = false;
 
         if (cache_params != nullptr && cache_params->mode != SD_CACHE_DISABLED) {
             bool percent_valid = true;
@@ -1674,7 +1728,7 @@ public:
                 } else {
                     EasyCacheConfig easycache_config;
                     easycache_config.enabled         = true;
-                    easycache_config.reuse_threshold = std::max(0.0f, cache_params->reuse_threshold);
+                    easycache_config.reuse_threshold = get_cache_reuse_threshold(*cache_params);
                     easycache_config.start_percent   = cache_params->start_percent;
                     easycache_config.end_percent     = cache_params->end_percent;
                     easycache_state.init(easycache_config, denoiser.get());
@@ -1695,7 +1749,7 @@ public:
                 } else {
                     UCacheConfig ucache_config;
                     ucache_config.enabled                = true;
-                    ucache_config.reuse_threshold        = std::max(0.0f, cache_params->reuse_threshold);
+                    ucache_config.reuse_threshold        = get_cache_reuse_threshold(*cache_params);
                     ucache_config.start_percent          = cache_params->start_percent;
                     ucache_config.end_percent            = cache_params->end_percent;
                     ucache_config.error_decay_rate       = std::max(0.0f, std::min(1.0f, cache_params->error_decay_rate));
@@ -1755,6 +1809,27 @@ public:
                         LOG_WARN("CacheDIT requested but could not be initialized for this run");
                     }
                 }
+            } else if (cache_params->mode == SD_CACHE_SPECTRUM) {
+                bool spectrum_supported = sd_version_is_unet(version) || sd_version_is_dit(version);
+                if (!spectrum_supported) {
+                    LOG_WARN("Spectrum requested but not supported for this model type (only UNET and DiT models)");
+                } else {
+                    SpectrumConfig spectrum_config;
+                    spectrum_config.w            = cache_params->spectrum_w;
+                    spectrum_config.m            = cache_params->spectrum_m;
+                    spectrum_config.lam          = cache_params->spectrum_lam;
+                    spectrum_config.window_size  = cache_params->spectrum_window_size;
+                    spectrum_config.flex_window  = cache_params->spectrum_flex_window;
+                    spectrum_config.warmup_steps = cache_params->spectrum_warmup_steps;
+                    spectrum_config.stop_percent = cache_params->spectrum_stop_percent;
+                    size_t total_steps           = sigmas.size() > 0 ? sigmas.size() - 1 : 0;
+                    spectrum_state.init(spectrum_config, total_steps);
+                    spectrum_enabled = true;
+                    LOG_INFO("Spectrum enabled - w: %.2f, m: %d, lam: %.2f, window: %d, flex: %.2f, warmup: %d, stop: %.0f%%",
+                             spectrum_config.w, spectrum_config.m, spectrum_config.lam,
+                             spectrum_config.window_size, spectrum_config.flex_window,
+                             spectrum_config.warmup_steps, spectrum_config.stop_percent * 100.0f);
+                }
             }
         }
 
@@ -1968,6 +2043,9 @@ public:
                 shifted_t             = std::max((int64_t)0, std::min((int64_t)(TIMESTEPS - 1), shifted_t));
                 LOG_DEBUG("shifting timestep from %.2f to %" PRId64 " (sigma: %.4f)", t, shifted_t, sigma);
                 timesteps_vec.assign(1, (float)shifted_t);
+            } else if (sd_version_is_anima(version)) {
+                // Anima uses normalized flow timesteps.
+                timesteps_vec.assign(1, t / static_cast<float>(TIMESTEPS));
             } else if (sd_version_is_z_image(version)) {
                 timesteps_vec.assign(1, 1000.f - t);
             } else {
@@ -1975,6 +2053,28 @@ public:
             }
 
             timesteps_vec = process_timesteps(timesteps_vec, init_latent, denoise_mask);
+
+            if (spectrum_enabled && spectrum_state.should_predict()) {
+                spectrum_state.predict(denoised);
+
+                if (denoise_mask != nullptr) {
+                    apply_mask(denoised, init_latent, denoise_mask);
+                }
+
+                if (sd_preview_cb != nullptr && sd_should_preview_denoised()) {
+                    if (step % sd_get_preview_interval() == 0) {
+                        preview_image(work_ctx, step, denoised, version, sd_preview_mode, preview_tensor, sd_preview_cb, sd_preview_cb_data, false);
+                    }
+                }
+
+                int64_t t1 = ggml_time_us();
+                if (step > 0 || step == -(int)steps) {
+                    int showstep = std::abs(step);
+                    pretty_progress(showstep, (int)steps, (t1 - t0) / 1000000.f / showstep);
+                }
+                return denoised;
+            }
+
             auto timesteps = vector_to_ggml_tensor(work_ctx, timesteps_vec);
             std::vector<float> guidance_vec(1, guidance.distilled_guidance);
             auto guidance_tensor = vector_to_ggml_tensor(work_ctx, guidance_vec);
@@ -2148,6 +2248,10 @@ public:
                 vec_denoised[i] = latent_result * c_out + vec_input[i] * c_skip;
             }
 
+            if (spectrum_enabled) {
+                spectrum_state.update(denoised);
+            }
+
             if (denoise_mask != nullptr) {
                 apply_mask(denoised, init_latent, denoise_mask);
             }
@@ -2239,6 +2343,14 @@ public:
             }
         }
 
+        if (spectrum_enabled && spectrum_state.total_steps_skipped > 0) {
+            size_t total_steps = sigmas.size() > 0 ? sigmas.size() - 1 : 0;
+            double speedup     = static_cast<double>(total_steps) /
+                             static_cast<double>(total_steps - spectrum_state.total_steps_skipped);
+            LOG_INFO("Spectrum skipped %d/%zu steps (%.2fx estimated speedup)",
+                     spectrum_state.total_steps_skipped, total_steps, speedup);
+        }
+
         if (inverse_noise_scaling) {
             x = denoiser->inverse_noise_scaling(sigmas[sigmas.size() - 1], x);
         }
@@ -2379,7 +2491,7 @@ public:
     }
 
     void process_latent_in(ggml_tensor* latent) {
-        if (sd_version_is_wan(version) || sd_version_is_qwen_image(version) || sd_version_is_flux2(version)) {
+        if (sd_version_is_wan(version) || sd_version_is_qwen_image(version) || sd_version_is_anima(version) || sd_version_is_flux2(version)) {
             int channel_dim = sd_version_is_flux2(version) ? 2 : 3;
             std::vector<float> latents_mean_vec;
             std::vector<float> latents_std_vec;
@@ -2418,7 +2530,7 @@ public:
     }
 
     void process_latent_out(ggml_tensor* latent) {
-        if (sd_version_is_wan(version) || sd_version_is_qwen_image(version) || sd_version_is_flux2(version)) {
+        if (sd_version_is_wan(version) || sd_version_is_qwen_image(version) || sd_version_is_anima(version) || sd_version_is_flux2(version)) {
             int channel_dim = sd_version_is_flux2(version) ? 2 : 3;
             std::vector<float> latents_mean_vec;
             std::vector<float> latents_std_vec;
@@ -2485,18 +2597,18 @@ public:
         tile_size_y = get_tile_size(params.tile_size_y, params.rel_size_y, latent_y);
     }
 
-    ggml_tensor* vae_encode(ggml_context* work_ctx, ggml_tensor* x, bool encode_video = false) {
+    ggml_tensor* vae_encode(ggml_context* work_ctx, ggml_tensor* x) {
         int64_t t0                 = ggml_time_ms();
         ggml_tensor* result        = nullptr;
         const int vae_scale_factor = get_vae_scale_factor();
         int64_t W                  = x->ne[0] / vae_scale_factor;
         int64_t H                  = x->ne[1] / vae_scale_factor;
         int64_t C                  = get_latent_channel();
-        if (vae_tiling_params.enabled && !encode_video) {
+        if (vae_tiling_params.enabled) {
             // TODO wan2.2 vae support?
             int64_t ne2;
             int64_t ne3;
-            if (sd_version_is_qwen_image(version)) {
+            if (sd_version_is_qwen_image(version) || sd_version_is_anima(version)) {
                 ne2 = 1;
                 ne3 = C * x->ne[3];
             } else {
@@ -2514,13 +2626,13 @@ public:
             result = ggml_new_tensor_4d(work_ctx, GGML_TYPE_F32, W, H, ne2, ne3);
         }
 
-        if (sd_version_is_qwen_image(version)) {
+        if (sd_version_is_qwen_image(version) || sd_version_is_anima(version)) {
             x = ggml_reshape_4d(work_ctx, x, x->ne[0], x->ne[1], 1, x->ne[2] * x->ne[3]);
         }
 
         if (!use_tiny_autoencoder) {
             process_vae_input_tensor(x);
-            if (vae_tiling_params.enabled && !encode_video) {
+            if (vae_tiling_params.enabled) {
                 float tile_overlap;
                 int tile_size_x, tile_size_y;
                 // multiply tile size for encode to keep the compute buffer size consistent
@@ -2529,20 +2641,20 @@ public:
                 LOG_DEBUG("VAE Tile size: %dx%d", tile_size_x, tile_size_y);
 
                 auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
-                    first_stage_model->compute(n_threads, in, false, &out, work_ctx);
+                    return first_stage_model->compute(n_threads, in, false, &out, work_ctx);
                 };
-                sd_tiling_non_square(x, result, vae_scale_factor, tile_size_x, tile_size_y, tile_overlap, on_tiling);
+                sd_tiling_non_square(x, result, vae_scale_factor, tile_size_x, tile_size_y, tile_overlap, circular_x, circular_y, on_tiling);
             } else {
                 first_stage_model->compute(n_threads, x, false, &result, work_ctx);
             }
             first_stage_model->free_compute_buffer();
         } else {
-            if (vae_tiling_params.enabled && !encode_video) {
+            if (vae_tiling_params.enabled) {
                 // split latent in 32x32 tiles and compute in several steps
                 auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
-                    tae_first_stage->compute(n_threads, in, false, &out, nullptr);
+                    return tae_first_stage->compute(n_threads, in, false, &out, nullptr);
                 };
-                sd_tiling(x, result, vae_scale_factor, 64, 0.5f, on_tiling);
+                sd_tiling(x, result, vae_scale_factor, 64, 0.5f, circular_x, circular_y, on_tiling);
             } else {
                 tae_first_stage->compute(n_threads, x, false, &result, work_ctx);
             }
@@ -2587,6 +2699,7 @@ public:
         ggml_tensor* latent;
         if (use_tiny_autoencoder ||
             sd_version_is_qwen_image(version) ||
+            sd_version_is_anima(version) ||
             sd_version_is_wan(version) ||
             sd_version_is_flux2(version) ||
             version == VERSION_CHROMA_RADIANCE) {
@@ -2603,17 +2716,17 @@ public:
         } else {
             latent = gaussian_latent_sample(work_ctx, vae_output);
         }
-        if (!use_tiny_autoencoder) {
+        if (!use_tiny_autoencoder && version != VERSION_SD1_PIX2PIX) {
             process_latent_in(latent);
         }
-        if (sd_version_is_qwen_image(version)) {
+        if (sd_version_is_qwen_image(version) || sd_version_is_anima(version)) {
             latent = ggml_reshape_4d(work_ctx, latent, latent->ne[0], latent->ne[1], latent->ne[3], 1);
         }
         return latent;
     }
 
-    ggml_tensor* encode_first_stage(ggml_context* work_ctx, ggml_tensor* x, bool encode_video = false) {
+    ggml_tensor* encode_first_stage(ggml_context* work_ctx, ggml_tensor* x) {
-        ggml_tensor* vae_output = vae_encode(work_ctx, x, encode_video);
+        ggml_tensor* vae_output = vae_encode(work_ctx, x);
         return get_first_stage_encoding(work_ctx, vae_output);
     }
 
@@ -2644,7 +2757,7 @@ public:
         }
         int64_t t0 = ggml_time_ms();
         if (!use_tiny_autoencoder) {
-            if (sd_version_is_qwen_image(version)) {
+            if (sd_version_is_qwen_image(version) || sd_version_is_anima(version)) {
                 x = ggml_reshape_4d(work_ctx, x, x->ne[0], x->ne[1], 1, x->ne[2] * x->ne[3]);
             }
             process_latent_out(x);
@@ -2658,11 +2771,15 @@ public:
 
                 // split latent in 32x32 tiles and compute in several steps
                 auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
-                    first_stage_model->compute(n_threads, in, true, &out, nullptr);
+                    return first_stage_model->compute(n_threads, in, true, &out, nullptr);
                 };
-                sd_tiling_non_square(x, result, vae_scale_factor, tile_size_x, tile_size_y, tile_overlap, on_tiling);
+                sd_tiling_non_square(x, result, vae_scale_factor, tile_size_x, tile_size_y, tile_overlap, circular_x, circular_y, on_tiling);
             } else {
-                first_stage_model->compute(n_threads, x, true, &result, work_ctx);
+                if (!first_stage_model->compute(n_threads, x, true, &result, work_ctx)) {
+                    LOG_ERROR("Failed to decode latetnts");
+                    first_stage_model->free_compute_buffer();
+                    return nullptr;
+                }
             }
             first_stage_model->free_compute_buffer();
             process_vae_output_tensor(result);
@@ -2670,11 +2787,15 @@ public:
             if (vae_tiling_params.enabled) {
                 // split latent in 64x64 tiles and compute in several steps
                 auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
-                    tae_first_stage->compute(n_threads, in, true, &out);
+                    return tae_first_stage->compute(n_threads, in, true, &out);
                 };
-                sd_tiling(x, result, vae_scale_factor, 64, 0.5f, on_tiling);
+                sd_tiling(x, result, vae_scale_factor, 64, 0.5f, circular_x, circular_y, on_tiling);
             } else {
-                tae_first_stage->compute(n_threads, x, true, &result);
+                if (!tae_first_stage->compute(n_threads, x, true, &result)) {
+                    LOG_ERROR("Failed to decode latetnts");
+                    tae_first_stage->free_compute_buffer();
+                    return nullptr;
+                }
             }
             tae_first_stage->free_compute_buffer();
         }
@@ -2684,6 +2805,16 @@ public:
         ggml_ext_tensor_clamp_inplace(result, 0.0f, 1.0f);
         return result;
     }
+
+    void set_flow_shift(float flow_shift = INFINITY) {
+        auto flow_denoiser = std::dynamic_pointer_cast<DiscreteFlowDenoiser>(denoiser);
+        if (flow_denoiser) {
+            if (flow_shift == INFINITY) {
+                flow_shift = default_flow_shift;
+            }
+            flow_denoiser->set_shift(flow_shift);
+        }
+    }
 };
 
 /*================================================= SD API ==================================================*/
@@ -2742,6 +2873,8 @@ const char* sample_method_to_str[] = {
     "lcm",
     "ddim_trailing",
     "tcd",
+    "res_multistep",
+    "res_2s",
 };
 
 const char* sd_sample_method_name(enum sample_method_t sample_method) {
@@ -2771,6 +2904,7 @@ const char* scheduler_to_str[] = {
     "smoothstep",
     "kl_optimal",
     "lcm",
+    "bong_tangent",
 };
 
 const char* sd_scheduler_name(enum scheduler_t scheduler) {
@@ -2862,7 +2996,7 @@ enum lora_apply_mode_t str_to_lora_apply_mode(const char* str) {
 void sd_cache_params_init(sd_cache_params_t* cache_params) {
     *cache_params                             = {};
     cache_params->mode                        = SD_CACHE_DISABLED;
-    cache_params->reuse_threshold             = 1.0f;
+    cache_params->reuse_threshold             = INFINITY;
     cache_params->start_percent               = 0.15f;
     cache_params->end_percent                 = 0.95f;
     cache_params->error_decay_rate            = 1.0f;
@@ -2878,6 +3012,13 @@ void sd_cache_params_init(sd_cache_params_t* cache_params) {
     cache_params->taylorseer_skip_interval    = 1;
     cache_params->scm_mask                    = nullptr;
     cache_params->scm_policy_dynamic          = true;
+    cache_params->spectrum_w                  = 0.40f;
+    cache_params->spectrum_m                  = 3;
+    cache_params->spectrum_lam                = 1.0f;
+    cache_params->spectrum_window_size        = 2;
+    cache_params->spectrum_flex_window        = 0.50f;
+    cache_params->spectrum_warmup_steps       = 4;
+    cache_params->spectrum_stop_percent       = 0.9f;
 }
 
 void sd_ctx_params_init(sd_ctx_params_t* sd_ctx_params) {
@@ -2901,7 +3042,6 @@ void sd_ctx_params_init(sd_ctx_params_t* sd_ctx_params) {
     sd_ctx_params->chroma_use_dit_mask     = true;
     sd_ctx_params->chroma_use_t5_mask      = false;
     sd_ctx_params->chroma_t5_mask_pad      = 1;
-    sd_ctx_params->flow_shift              = INFINITY;
 }
 
 char* sd_ctx_params_to_str(const sd_ctx_params_t* sd_ctx_params) {
@@ -2936,6 +3076,7 @@ char* sd_ctx_params_to_str(const sd_ctx_params_t* sd_ctx_params) {
              "keep_clip_on_cpu: %s\n"
              "keep_control_net_on_cpu: %s\n"
              "keep_vae_on_cpu: %s\n"
+             "flash_attn: %s\n"
              "diffusion_flash_attn: %s\n"
              "circular_x: %s\n"
              "circular_y: %s\n"
@@ -2967,6 +3108,7 @@ char* sd_ctx_params_to_str(const sd_ctx_params_t* sd_ctx_params) {
              BOOL_STR(sd_ctx_params->keep_clip_on_cpu),
              BOOL_STR(sd_ctx_params->keep_control_net_on_cpu),
              BOOL_STR(sd_ctx_params->keep_vae_on_cpu),
+             BOOL_STR(sd_ctx_params->flash_attn),
              BOOL_STR(sd_ctx_params->diffusion_flash_attn),
              BOOL_STR(sd_ctx_params->circular_x),
              BOOL_STR(sd_ctx_params->circular_y),
@@ -2991,6 +3133,7 @@ void sd_sample_params_init(sd_sample_params_t* sample_params) {
     sample_params->sample_steps                = 20;
     sample_params->custom_sigmas               = nullptr;
     sample_params->custom_sigmas_count         = 0;
+    sample_params->flow_shift                  = INFINITY;
 }
 
 char* sd_sample_params_to_str(const sd_sample_params_t* sample_params) {
@@ -3011,7 +3154,8 @@ char* sd_sample_params_to_str(const sd_sample_params_t* sample_params) {
              "sample_method: %s, "
              "sample_steps: %d, "
              "eta: %.2f, "
-             "shifted_timestep: %d)",
+             "shifted_timestep: %d, "
+             "flow_shift: %.2f)",
              sample_params->guidance.txt_cfg,
              std::isfinite(sample_params->guidance.img_cfg)
                  ? sample_params->guidance.img_cfg
@@ -3025,7 +3169,8 @@ char* sd_sample_params_to_str(const sd_sample_params_t* sample_params) {
              sd_sample_method_name(sample_params->sample_method),
              sample_params->sample_steps,
              sample_params->eta,
-             sample_params->shifted_timestep);
+             sample_params->shifted_timestep,
+             sample_params->flow_shift);
 
     return buf;
 }
@@ -3097,7 +3242,7 @@ char* sd_img_gen_params_to_str(const sd_img_gen_params_t* sd_img_gen_params) {
     snprintf(buf + strlen(buf), 4096 - strlen(buf),
              "cache: %s (threshold=%.3f, start=%.2f, end=%.2f)\n",
              cache_mode_str,
-             sd_img_gen_params->cache.reuse_threshold,
+             get_cache_reuse_threshold(sd_img_gen_params->cache),
              sd_img_gen_params->cache.start_percent,
              sd_img_gen_params->cache.end_percent);
     free(sample_params_str);
@@ -3439,6 +3584,7 @@ sd_image_t* generate_image_internal(sd_ctx_t* sd_ctx,
         ggml_free(work_ctx);
         return nullptr;
     }
+    memset(result_images, 0, batch_count * sizeof(sd_image_t));
 
     for (size_t i = 0; i < decoded_images.size(); i++) {
         result_images[i].width   = width;
@@ -3453,6 +3599,7 @@ sd_image_t* generate_image_internal(sd_ctx_t* sd_ctx,
 
 sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_gen_params) {
     sd_ctx->sd->vae_tiling_params = sd_img_gen_params->vae_tiling_params;
+
     int width  = sd_img_gen_params->width;
     int height = sd_img_gen_params->height;
 
@@ -3468,6 +3615,40 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
         LOG_WARN("align up %dx%d to %dx%d (multiple=%d)", sd_img_gen_params->width, sd_img_gen_params->height, width, height, spatial_multiple);
     }
 
+    bool circular_x = sd_ctx->sd->circular_x;
+    bool circular_y = sd_ctx->sd->circular_y;
+
+    if (!sd_img_gen_params->vae_tiling_params.enabled) {
+        if (sd_ctx->sd->first_stage_model) {
+            sd_ctx->sd->first_stage_model->set_circular_axes(sd_ctx->sd->circular_x, sd_ctx->sd->circular_y);
+        }
+        if (sd_ctx->sd->tae_first_stage) {
+            sd_ctx->sd->tae_first_stage->set_circular_axes(sd_ctx->sd->circular_x, sd_ctx->sd->circular_y);
+        }
+    } else {
+        int tile_size_x, tile_size_y;
+        float _overlap;
+        int latent_size_x = width / sd_ctx->sd->get_vae_scale_factor();
+        int latent_size_y = height / sd_ctx->sd->get_vae_scale_factor();
+        sd_ctx->sd->get_tile_sizes(tile_size_x, tile_size_y, _overlap, sd_img_gen_params->vae_tiling_params, latent_size_x, latent_size_y);
+
+        // force disable circular padding for vae if tiling is enabled unless latent is smaller than tile size
+        // otherwise it will cause artifacts at the edges of the tiles
+        sd_ctx->sd->circular_x = sd_ctx->sd->circular_x && (tile_size_x >= latent_size_x);
+        sd_ctx->sd->circular_y = sd_ctx->sd->circular_y && (tile_size_y >= latent_size_y);
+
+        if (sd_ctx->sd->first_stage_model) {
+            sd_ctx->sd->first_stage_model->set_circular_axes(sd_ctx->sd->circular_x, sd_ctx->sd->circular_y);
+        }
+        if (sd_ctx->sd->tae_first_stage) {
+            sd_ctx->sd->tae_first_stage->set_circular_axes(sd_ctx->sd->circular_x, sd_ctx->sd->circular_y);
+        }
+
+        // disable circular tiling if it's enabled for the VAE
+        sd_ctx->sd->circular_x = circular_x && (tile_size_x < latent_size_x);
+        sd_ctx->sd->circular_y = circular_y && (tile_size_y < latent_size_y);
+    }
+
     LOG_DEBUG("generate_image %dx%d", width, height);
     if (sd_ctx == nullptr || sd_img_gen_params == nullptr) {
         return nullptr;
@@ -3495,6 +3676,8 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
 
     size_t t0 = ggml_time_ms();
 
+    sd_ctx->sd->set_flow_shift(sd_img_gen_params->sample_params.flow_shift);
+
     // Apply lora
     sd_ctx->sd->apply_loras(sd_img_gen_params->loras, sd_img_gen_params->lora_count);
 
@@ -3735,6 +3918,10 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
                                                         denoise_mask,
                                                         &sd_img_gen_params->cache);
 
+    // restore circular params
+    sd_ctx->sd->circular_x = circular_x;
+    sd_ctx->sd->circular_y = circular_y;
+
     size_t t2 = ggml_time_ms();
 
     LOG_INFO("generate_image completed in %.2fs", (t2 - t0) * 1.0f / 1000);
@@ -3770,6 +3957,8 @@ SD_API sd_image_t* generate_video(sd_ctx_t* sd_ctx, const sd_vid_gen_params_t* s
     }
     LOG_INFO("generate_video %dx%dx%d", width, height, frames);
 
+    sd_ctx->sd->set_flow_shift(sd_vid_gen_params->sample_params.flow_shift);
+
     enum sample_method_t sample_method = sd_vid_gen_params->sample_params.sample_method;
     if (sample_method == SAMPLE_METHOD_COUNT) {
         sample_method = sd_get_default_sample_method(sd_ctx);

src/t5.hpp

@@ -14,6 +14,7 @@
 #include "ggml_extend.hpp"
 #include "json.hpp"
 #include "model.h"
+#include "vocab/vocab.h"
 
 // Port from: https://github.com/google/sentencepiece/blob/master/src/unigram_model.h
 // and https://github.com/google/sentencepiece/blob/master/src/unigram_model.h.
@@ -341,9 +342,9 @@ protected:
 public:
     explicit T5UniGramTokenizer(bool is_umt5 = false) {
         if (is_umt5) {
-            InitializePieces(ModelLoader::load_umt5_tokenizer_json());
+            InitializePieces(load_umt5_tokenizer_json());
         } else {
-            InitializePieces(ModelLoader::load_t5_tokenizer_json());
+            InitializePieces(load_t5_tokenizer_json());
         }
 
         min_score_ = FLT_MAX;
@@ -515,7 +516,7 @@ public:
         auto wi_1 = std::dynamic_pointer_cast<Linear>(blocks["wi_1"]);
         auto wo   = std::dynamic_pointer_cast<Linear>(blocks["wo"]);
 
-        auto hidden_gelu   = ggml_gelu_inplace(ctx->ggml_ctx, wi_0->forward(ctx, x));
+        auto hidden_gelu   = ggml_ext_gelu(ctx->ggml_ctx, wi_0->forward(ctx, x), true);
         auto hidden_linear = wi_1->forward(ctx, x);
         x                  = ggml_mul_inplace(ctx->ggml_ctx, hidden_gelu, hidden_linear);
         x                  = wo->forward(ctx, x);
@@ -608,7 +609,7 @@ public:
             }
         }
 
-        k = ggml_scale_inplace(ctx->ggml_ctx, k, ::sqrtf(static_cast<float>(d_head)));
+        k = ggml_ext_scale(ctx->ggml_ctx, k, ::sqrtf(static_cast<float>(d_head)), true);
 
         x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, num_heads, mask);  // [N, n_token, d_head * n_head]
 

src/tae.hpp

@@ -17,22 +17,43 @@ class TAEBlock : public UnaryBlock {
 protected:
     int n_in;
     int n_out;
+    bool use_midblock_gn;
 
 public:
-    TAEBlock(int n_in, int n_out)
+    TAEBlock(int n_in, int n_out, bool use_midblock_gn = false)
-        : n_in(n_in), n_out(n_out) {
+        : n_in(n_in), n_out(n_out), use_midblock_gn(use_midblock_gn) {
         blocks["conv.0"] = std::shared_ptr<GGMLBlock>(new Conv2d(n_in, n_out, {3, 3}, {1, 1}, {1, 1}));
         blocks["conv.2"] = std::shared_ptr<GGMLBlock>(new Conv2d(n_out, n_out, {3, 3}, {1, 1}, {1, 1}));
         blocks["conv.4"] = std::shared_ptr<GGMLBlock>(new Conv2d(n_out, n_out, {3, 3}, {1, 1}, {1, 1}));
         if (n_in != n_out) {
             blocks["skip"] = std::shared_ptr<GGMLBlock>(new Conv2d(n_in, n_out, {1, 1}, {1, 1}, {1, 1}, {1, 1}, false));
         }
+        if (use_midblock_gn) {
+            int n_gn         = n_in * 4;
+            blocks["pool.0"] = std::shared_ptr<GGMLBlock>(new Conv2d(n_in, n_gn, {1, 1}, {1, 1}, {0, 0}, {1, 1}, false));
+            blocks["pool.1"] = std::shared_ptr<GGMLBlock>(new GroupNorm(4, n_gn));
+            // pool.2 is ReLU, handled in forward
+            blocks["pool.3"] = std::shared_ptr<GGMLBlock>(new Conv2d(n_gn, n_in, {1, 1}, {1, 1}, {0, 0}, {1, 1}, false));
+        }
     }
 
     struct ggml_tensor* forward(GGMLRunnerContext* ctx, struct ggml_tensor* x) override {
         // x: [n, n_in, h, w]
         // return: [n, n_out, h, w]
 
+        if (use_midblock_gn) {
+            auto pool_0 = std::dynamic_pointer_cast<Conv2d>(blocks["pool.0"]);
+            auto pool_1 = std::dynamic_pointer_cast<GroupNorm>(blocks["pool.1"]);
+            auto pool_3 = std::dynamic_pointer_cast<Conv2d>(blocks["pool.3"]);
+
+            auto p = pool_0->forward(ctx, x);
+            p      = pool_1->forward(ctx, p);
+            p      = ggml_relu_inplace(ctx->ggml_ctx, p);
+            p      = pool_3->forward(ctx, p);
+
+            x = ggml_add(ctx->ggml_ctx, x, p);
+        }
+
         auto conv_0 = std::dynamic_pointer_cast<Conv2d>(blocks["conv.0"]);
         auto conv_2 = std::dynamic_pointer_cast<Conv2d>(blocks["conv.2"]);
         auto conv_4 = std::dynamic_pointer_cast<Conv2d>(blocks["conv.4"]);
@@ -62,7 +83,7 @@ class TinyEncoder : public UnaryBlock {
     int num_blocks  = 3;
 
 public:
-    TinyEncoder(int z_channels = 4)
+    TinyEncoder(int z_channels = 4, bool use_midblock_gn = false)
         : z_channels(z_channels) {
         int index                       = 0;
         blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new Conv2d(in_channels, channels, {3, 3}, {1, 1}, {1, 1}));
@@ -80,7 +101,7 @@ public:
 
         blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new Conv2d(channels, channels, {3, 3}, {2, 2}, {1, 1}, {1, 1}, false));
         for (int i = 0; i < num_blocks; i++) {
-            blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new TAEBlock(channels, channels));
+            blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new TAEBlock(channels, channels, use_midblock_gn));
         }
 
         blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new Conv2d(channels, z_channels, {3, 3}, {1, 1}, {1, 1}));
@@ -107,7 +128,7 @@ class TinyDecoder : public UnaryBlock {
     int num_blocks   = 3;
 
 public:
-    TinyDecoder(int z_channels = 4)
+    TinyDecoder(int z_channels = 4, bool use_midblock_gn = false)
         : z_channels(z_channels) {
         int index = 0;
 
@@ -115,7 +136,7 @@ public:
         index++;  // nn.ReLU()
 
         for (int i = 0; i < num_blocks; i++) {
-            blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new TAEBlock(channels, channels));
+            blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new TAEBlock(channels, channels, use_midblock_gn));
         }
         index++;  // nn.Upsample()
         blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new Conv2d(channels, channels, {3, 3}, {1, 1}, {1, 1}, {1, 1}, false));
@@ -140,9 +161,9 @@ public:
         // z: [n, z_channels, h, w]
         // return: [n, out_channels, h*8, w*8]
 
-        auto h = ggml_scale(ctx->ggml_ctx, z, 1.0f / 3.0f);
+        auto h = ggml_ext_scale(ctx->ggml_ctx, z, 1.0f / 3.0f);
         h      = ggml_tanh_inplace(ctx->ggml_ctx, h);
-        h      = ggml_scale(ctx->ggml_ctx, h, 3.0f);
+        h      = ggml_ext_scale(ctx->ggml_ctx, h, 3.0f);
 
         for (int i = 0; i < num_blocks * 3 + 10; i++) {
             if (blocks.find(std::to_string(i)) == blocks.end()) {
@@ -379,10 +400,11 @@ public:
         auto first_conv = std::dynamic_pointer_cast<Conv2d>(blocks["1"]);
 
         // Clamp()
-        auto h = ggml_scale_inplace(ctx->ggml_ctx,
+        auto h = ggml_ext_scale(ctx->ggml_ctx,
                                 ggml_tanh_inplace(ctx->ggml_ctx,
-                                                      ggml_scale(ctx->ggml_ctx, z, 1.0f / 3.0f)),
+                                                  ggml_ext_scale(ctx->ggml_ctx, z, 1.0f / 3.0f)),
-                                    3.0f);
+                                3.0f,
+                                true);
 
         h         = first_conv->forward(ctx, h);
         h         = ggml_relu_inplace(ctx->ggml_ctx, h);
@@ -470,29 +492,44 @@ public:
 class TAESD : public GGMLBlock {
 protected:
     bool decode_only;
+    bool taef2 = false;
 
 public:
     TAESD(bool decode_only = true, SDVersion version = VERSION_SD1)
         : decode_only(decode_only) {
         int z_channels       = 4;
+        bool use_midblock_gn = false;
+        taef2                = sd_version_is_flux2(version);
+
         if (sd_version_is_dit(version)) {
             z_channels = 16;
         }
-        blocks["decoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyDecoder(z_channels));
+        if (taef2) {
+            z_channels      = 32;
+            use_midblock_gn = true;
+        }
+        blocks["decoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyDecoder(z_channels, use_midblock_gn));
 
         if (!decode_only) {
-            blocks["encoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyEncoder(z_channels));
+            blocks["encoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyEncoder(z_channels, use_midblock_gn));
         }
     }
 
     struct ggml_tensor* decode(GGMLRunnerContext* ctx, struct ggml_tensor* z) {
         auto decoder = std::dynamic_pointer_cast<TinyDecoder>(blocks["decoder.layers"]);
+        if (taef2) {
+            z = unpatchify(ctx->ggml_ctx, z, 2);
+        }
         return decoder->forward(ctx, z);
     }
 
     struct ggml_tensor* encode(GGMLRunnerContext* ctx, struct ggml_tensor* x) {
         auto encoder = std::dynamic_pointer_cast<TinyEncoder>(blocks["encoder.layers"]);
-        return encoder->forward(ctx, x);
+        auto z       = encoder->forward(ctx, x);
+        if (taef2) {
+            z = patchify(ctx->ggml_ctx, z, 2);
+        }
+        return z;
     }
 };
 

src/tokenize_util.cpp

@@ -919,17 +919,23 @@ std::vector<std::string> token_split(const std::string& text) {
 
         // `\s*[\r\n]+|\s+(?!\S)|\s+`
         if (is_space(cp)) {
-            std::string token = codepoint_to_utf8(cp);
+            std::string token;
-            ++i;
+            bool saw_new_line = false;
 
             while (i < cps.size() && is_space(cps[i])) {
                 token += codepoint_to_utf8(cps[i]);
-                ++i;
+
                 if (cps[i] == U'\r' || cps[i] == U'\n') {
+                    saw_new_line = true;
+                } else {
+                    if (saw_new_line) {
                         break;
                     }
                 }
 
+                ++i;
+            }
+
             tokens.push_back(token);
             continue;
         }

src/ucache.hpp

@@ -19,6 +19,7 @@ struct UCacheConfig {
     bool adaptive_threshold     = true;
     float early_step_multiplier = 0.5f;
     float late_step_multiplier  = 1.5f;
+    float relative_norm_gain    = 1.6f;
     bool reset_error_on_compute = true;
 };
 
@@ -45,14 +46,16 @@ struct UCacheState {
     bool has_output_prev_norm             = false;
     bool has_relative_transformation_rate = false;
     float relative_transformation_rate    = 0.0f;
-    float cumulative_change_rate          = 0.0f;
     float last_input_change               = 0.0f;
     bool has_last_input_change            = false;
+    float output_change_ema               = 0.0f;
+    bool has_output_change_ema            = false;
     int total_steps_skipped               = 0;
     int current_step_index                = -1;
     int steps_computed_since_active       = 0;
+    int expected_total_steps              = 0;
+    int consecutive_skipped_steps         = 0;
     float accumulated_error               = 0.0f;
-    float reference_output_norm           = 0.0f;
 
     struct BlockMetrics {
         float sum_transformation_rate = 0.0f;
@@ -106,14 +109,16 @@ struct UCacheState {
         has_output_prev_norm             = false;
         has_relative_transformation_rate = false;
         relative_transformation_rate     = 0.0f;
-        cumulative_change_rate           = 0.0f;
         last_input_change                = 0.0f;
         has_last_input_change            = false;
+        output_change_ema                = 0.0f;
+        has_output_change_ema            = false;
         total_steps_skipped              = 0;
         current_step_index               = -1;
         steps_computed_since_active      = 0;
+        expected_total_steps             = 0;
+        consecutive_skipped_steps        = 0;
         accumulated_error                = 0.0f;
-        reference_output_norm            = 0.0f;
         block_metrics.reset();
         total_active_steps = 0;
     }
@@ -134,6 +139,7 @@ struct UCacheState {
             return;
         }
         size_t n_steps       = sigmas.size() - 1;
+        expected_total_steps = static_cast<int>(n_steps);
 
         size_t start_step = static_cast<size_t>(config.start_percent * n_steps);
         size_t end_step   = static_cast<size_t>(config.end_percent * n_steps);
@@ -207,11 +213,15 @@ struct UCacheState {
         }
 
         int effective_total = estimated_total_steps;
+        if (effective_total <= 0) {
+            effective_total = expected_total_steps;
+        }
         if (effective_total <= 0) {
             effective_total = std::max(20, steps_computed_since_active * 2);
         }
 
         float progress = (effective_total > 0) ? (static_cast<float>(steps_computed_since_active) / effective_total) : 0.0f;
+        progress       = std::max(0.0f, std::min(1.0f, progress));
 
         float multiplier = 1.0f;
         if (progress < 0.2f) {
@@ -309,17 +319,31 @@ struct UCacheState {
 
         if (has_output_prev_norm && has_relative_transformation_rate &&
             last_input_change > 0.0f && output_prev_norm > 0.0f) {
-            float approx_output_change_rate = (relative_transformation_rate * last_input_change) / output_prev_norm;
+            float approx_output_change = relative_transformation_rate * last_input_change;
+            float approx_output_change_rate;
+            if (config.use_relative_threshold) {
+                float base_scale          = std::max(output_prev_norm, 1e-6f);
+                float dyn_scale           = has_output_change_ema
+                                                ? std::max(output_change_ema * std::max(1.0f, config.relative_norm_gain), 1e-6f)
+                                                : base_scale;
+                float scale               = std::sqrt(base_scale * dyn_scale);
+                approx_output_change_rate = approx_output_change / scale;
+            } else {
+                approx_output_change_rate = approx_output_change;
+            }
+            // Increase estimated error with skip horizon to avoid long extrapolation streaks
+            approx_output_change_rate *= (1.0f + 0.50f * consecutive_skipped_steps);
             accumulated_error = accumulated_error * config.error_decay_rate + approx_output_change_rate;
 
             float effective_threshold = get_adaptive_threshold();
-            if (config.use_relative_threshold && reference_output_norm > 0.0f) {
+            if (!config.use_relative_threshold && output_prev_norm > 0.0f) {
-                effective_threshold = effective_threshold * reference_output_norm;
+                effective_threshold = effective_threshold * output_prev_norm;
             }
 
             if (accumulated_error < effective_threshold) {
                 skip_current_step = true;
                 total_steps_skipped++;
+                consecutive_skipped_steps++;
                 apply_cache(cond, input, output);
                 return true;
             } else if (config.reset_error_on_compute) {
@@ -340,6 +364,8 @@ struct UCacheState {
         if (cond != anchor_condition) {
             return;
         }
+        steps_computed_since_active++;
+        consecutive_skipped_steps = 0;
 
         size_t ne      = static_cast<size_t>(ggml_nelements(input));
         float* in_data = (float*)input->data;
@@ -359,6 +385,14 @@ struct UCacheState {
                 output_change /= static_cast<float>(ne);
             }
         }
+        if (std::isfinite(output_change) && output_change > 0.0f) {
+            if (!has_output_change_ema) {
+                output_change_ema     = output_change;
+                has_output_change_ema = true;
+            } else {
+                output_change_ema = 0.8f * output_change_ema + 0.2f * output_change;
+            }
+        }
 
         prev_output.resize(ne);
         for (size_t i = 0; i < ne; ++i) {
@@ -373,10 +407,6 @@ struct UCacheState {
         output_prev_norm     = (ne > 0) ? (mean_abs / static_cast<float>(ne)) : 0.0f;
         has_output_prev_norm = output_prev_norm > 0.0f;
 
-        if (reference_output_norm == 0.0f) {
-            reference_output_norm = output_prev_norm;
-        }
-
         if (has_last_input_change && last_input_change > 0.0f && output_change > 0.0f) {
             float rate = output_change / last_input_change;
             if (std::isfinite(rate)) {

src/unet.hpp

@@ -1,8 +1,7 @@
 #ifndef __UNET_HPP__
 #define __UNET_HPP__
 
-#include "common.hpp"
+#include "common_block.hpp"
-#include "ggml_extend.hpp"
 #include "model.h"
 
 /*==================================================== UnetModel =====================================================*/
@@ -201,6 +200,9 @@ public:
             num_head_channels     = 64;
             num_heads             = -1;
             use_linear_projection = true;
+            if (version == VERSION_SDXL_VEGA) {
+                transformer_depth = {1, 1, 2};
+            }
         } else if (version == VERSION_SVD) {
             in_channels           = 8;
             out_channels          = 4;
@@ -319,7 +321,7 @@ public:
         }
         if (!tiny_unet) {
             blocks["middle_block.0"] = std::shared_ptr<GGMLBlock>(get_resblock(ch, time_embed_dim, ch));
-            if (version != VERSION_SDXL_SSD1B) {
+            if (version != VERSION_SDXL_SSD1B && version != VERSION_SDXL_VEGA) {
                 blocks["middle_block.1"] = std::shared_ptr<GGMLBlock>(get_attention_layer(ch,
                                                                                           n_head,
                                                                                           d_head,
@@ -520,13 +522,13 @@ public:
         // middle_block
         if (!tiny_unet) {
             h = resblock_forward("middle_block.0", ctx, h, emb, num_video_frames);  // [N, 4*model_channels, h/8, w/8]
-            if (version != VERSION_SDXL_SSD1B) {
+            if (version != VERSION_SDXL_SSD1B && version != VERSION_SDXL_VEGA) {
                 h = attention_layer_forward("middle_block.1", ctx, h, context, num_video_frames);  // [N, 4*model_channels, h/8, w/8]
                 h = resblock_forward("middle_block.2", ctx, h, emb, num_video_frames);             // [N, 4*model_channels, h/8, w/8]
             }
         }
         if (controls.size() > 0) {
-            auto cs = ggml_scale_inplace(ctx->ggml_ctx, controls[controls.size() - 1], control_strength);
+            auto cs = ggml_ext_scale(ctx->ggml_ctx, controls[controls.size() - 1], control_strength, true);
             h       = ggml_add(ctx->ggml_ctx, h, cs);  // middle control
         }
         int control_offset = static_cast<int>(controls.size() - 2);
@@ -539,7 +541,7 @@ public:
                 hs.pop_back();
 
                 if (controls.size() > 0) {
-                    auto cs = ggml_scale_inplace(ctx->ggml_ctx, controls[control_offset], control_strength);
+                    auto cs = ggml_ext_scale(ctx->ggml_ctx, controls[control_offset], control_strength, true);
                     h_skip  = ggml_add(ctx->ggml_ctx, h_skip, cs);  // control net condition
                     control_offset--;
                 }

src/upscaler.cpp

@@ -89,10 +89,11 @@ struct UpscalerGGML {
 
         ggml_tensor* upscaled = ggml_new_tensor_4d(upscale_ctx, GGML_TYPE_F32, output_width, output_height, 3, 1);
         auto on_tiling        = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
-            esrgan_upscaler->compute(n_threads, in, &out);
+            return esrgan_upscaler->compute(n_threads, in, &out);
         };
         int64_t t0 = ggml_time_ms();
-        sd_tiling(input_image_tensor, upscaled, esrgan_upscaler->scale, esrgan_upscaler->tile_size, 0.25f, on_tiling);
+        // TODO: circular upscaling?
+        sd_tiling(input_image_tensor, upscaled, esrgan_upscaler->scale, esrgan_upscaler->tile_size, 0.25f, false, false, on_tiling);
         esrgan_upscaler->free_compute_buffer();
         ggml_ext_tensor_clamp_inplace(upscaled, 0.f, 1.f);
         uint8_t* upscaled_data = ggml_tensor_to_sd_image(upscaled);

src/vae.hpp

@@ -1,8 +1,7 @@
 #ifndef __VAE_HPP__
 #define __VAE_HPP__
 
-#include "common.hpp"
+#include "common_block.hpp"
-#include "ggml_extend.hpp"
 
 /*================================================== AutoEncoderKL ===================================================*/
 
@@ -141,7 +140,7 @@ public:
             v = ggml_reshape_3d(ctx->ggml_ctx, v, c, h * w, n);                        // [N, h * w, in_channels]
         }
 
-        h_ = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, 1, nullptr, false, true, false);
+        h_ = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, 1, nullptr, false, ctx->flash_attn_enabled);
 
         if (use_linear) {
             h_ = proj_out->forward(ctx, h_);  // [N, h * w, in_channels]
@@ -253,8 +252,8 @@ public:
 
         float alpha = get_alpha();
         x           = ggml_add(ctx->ggml_ctx,
-                               ggml_scale(ctx->ggml_ctx, x, alpha),
+                               ggml_ext_scale(ctx->ggml_ctx, x, alpha),
-                               ggml_scale(ctx->ggml_ctx, x_mix, 1.0f - alpha));
+                               ggml_ext_scale(ctx->ggml_ctx, x_mix, 1.0f - alpha));
 
         x = ggml_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, x, 0, 2, 1, 3));  // b c t (h w) -> b t c (h w)
         x = ggml_reshape_4d(ctx->ggml_ctx, x, W, H, C, T * B);                     // b t c (h w) -> (b t) c h w

src/vocab/clip_t5.hpp

@@ -1,4 +1,4 @@
-static unsigned char merges_utf8_c_str[] = {
+static const unsigned char clip_merges_utf8_c_str[] = {
     0x23,
     0x76,
     0x65,
@@ -524620,7 +524620,7 @@ static unsigned char merges_utf8_c_str[] = {
     0x0a,
 };
 
-static unsigned char t5_tokenizer_json_str[] = {
+static const unsigned char t5_tokenizer_json_str[] = {
     0x7b,
     0x0a,
     0x20,

src/vocab/mistral.hpp

@@ -1,4 +1,4 @@
-unsigned char mistral_merges_utf8_c_str[] = {
+static const unsigned char mistral_merges_utf8_c_str[] = {
   0xc4, 0xa0, 0x20, 0xc4, 0xa0, 0x0a, 0xc4, 0xa0, 0x20, 0x74, 0x0a, 0x65,
   0x20, 0x72, 0x0a, 0x69, 0x20, 0x6e, 0x0a, 0xc4, 0xa0, 0x20, 0xc4, 0xa0,
   0xc4, 0xa0, 0xc4, 0xa0, 0x0a, 0xc4, 0xa0, 0xc4, 0xa0, 0x20, 0xc4, 0xa0,
@@ -260614,7 +260614,7 @@ unsigned char mistral_merges_utf8_c_str[] = {
   0xc3, 0xa5, 0xc4, 0xb2, 0xc4, 0xb0, 0x20, 0xc3, 0xa6, 0xc2, 0xb1, 0xc4,
   0xab, 0xc3, 0xa4, 0xc2, 0xb9, 0xc2, 0xa6, 0x0a,
 };
-unsigned char mistral_vocab_json_utf8_c_str[] = {
+static const unsigned char mistral_vocab_json_utf8_c_str[] = {
   0x7b, 0x22, 0x3c, 0x75, 0x6e, 0x6b, 0x3e, 0x22, 0x3a, 0x20, 0x30, 0x2c,
   0x20, 0x22, 0x3c, 0x73, 0x3e, 0x22, 0x3a, 0x20, 0x31, 0x2c, 0x20, 0x22,
   0x3c, 0x2f, 0x73, 0x3e, 0x22, 0x3a, 0x20, 0x32, 0x2c, 0x20, 0x22, 0x5b,

src/vocab/qwen.hpp

@@ -1,4 +1,4 @@
-unsigned char qwen2_merges_utf8_c_str[] = {
+static const unsigned char qwen2_merges_utf8_c_str[] = {
   0xc4, 0xa0, 0x20, 0xc4, 0xa0, 0x0a, 0xc4, 0xa0, 0xc4, 0xa0, 0x20, 0xc4,
   0xa0, 0xc4, 0xa0, 0x0a, 0x69, 0x20, 0x6e, 0x0a, 0xc4, 0xa0, 0x20, 0x74,
   0x0a, 0xc4, 0xa0, 0xc4, 0xa0, 0xc4, 0xa0, 0xc4, 0xa0, 0x20, 0xc4, 0xa0,

src/vocab/umt5.hpp

@@ -1,4 +1,4 @@
-unsigned char umt5_tokenizer_json_str[] = {
+static const unsigned char umt5_tokenizer_json_str[] = {
   0x7b, 0x22, 0x76, 0x65, 0x72, 0x73, 0x69, 0x6f, 0x6e, 0x22, 0x3a, 0x20,
   0x22, 0x31, 0x2e, 0x30, 0x22, 0x2c, 0x20, 0x22, 0x74, 0x72, 0x75, 0x6e,
   0x63, 0x61, 0x74, 0x69, 0x6f, 0x6e, 0x22, 0x3a, 0x20, 0x6e, 0x75, 0x6c,

src/vocab/vocab.cpp

@@ -0,0 +1,35 @@
+#include "vocab.h"
+#include "clip_t5.hpp"
+#include "mistral.hpp"
+#include "qwen.hpp"
+#include "umt5.hpp"
+
+std::string load_clip_merges() {
+    std::string merges_utf8_str(reinterpret_cast<const char*>(clip_merges_utf8_c_str), sizeof(clip_merges_utf8_c_str));
+    return merges_utf8_str;
+}
+
+std::string load_qwen2_merges() {
+    std::string merges_utf8_str(reinterpret_cast<const char*>(qwen2_merges_utf8_c_str), sizeof(qwen2_merges_utf8_c_str));
+    return merges_utf8_str;
+}
+
+std::string load_mistral_merges() {
+    std::string merges_utf8_str(reinterpret_cast<const char*>(mistral_merges_utf8_c_str), sizeof(mistral_merges_utf8_c_str));
+    return merges_utf8_str;
+}
+
+std::string load_mistral_vocab_json() {
+    std::string json_str(reinterpret_cast<const char*>(mistral_vocab_json_utf8_c_str), sizeof(mistral_vocab_json_utf8_c_str));
+    return json_str;
+}
+
+std::string load_t5_tokenizer_json() {
+    std::string json_str(reinterpret_cast<const char*>(t5_tokenizer_json_str), sizeof(t5_tokenizer_json_str));
+    return json_str;
+}
+
+std::string load_umt5_tokenizer_json() {
+    std::string json_str(reinterpret_cast<const char*>(umt5_tokenizer_json_str), sizeof(umt5_tokenizer_json_str));
+    return json_str;
+}

src/vocab/vocab.h

@@ -0,0 +1,13 @@
+#ifndef __VOCAB_H__
+#define __VOCAB_H__
+
+#include <string>
+
+std::string load_clip_merges();
+std::string load_qwen2_merges();
+std::string load_mistral_merges();
+std::string load_mistral_vocab_json();
+std::string load_t5_tokenizer_json();
+std::string load_umt5_tokenizer_json();
+
+#endif  // __VOCAB_H__

src/wan.hpp

@@ -5,9 +5,8 @@
 #include <memory>
 #include <utility>
 
-#include "common.hpp"
+#include "common_block.hpp"
 #include "flux.hpp"
-#include "ggml_extend.hpp"
 #include "rope.hpp"
 #include "vae.hpp"
 
@@ -573,7 +572,7 @@ namespace WAN {
             v      = ggml_reshape_3d(ctx->ggml_ctx, v, h * w, c, n);  // [t, c, h * w]
 
             v = ggml_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, v, 1, 0, 2, 3));                            // [t, h * w, c]
-            x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, 1, nullptr, false, true, false);  // [t, h * w, c]
+            x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, 1, nullptr, false, ctx->flash_attn_enabled);  // [t, h * w, c]
 
             x = ggml_ext_cont(ctx->ggml_ctx, ggml_permute(ctx->ggml_ctx, x, 1, 0, 2, 3));  // [t, c, h * w]
             x = ggml_reshape_4d(ctx->ggml_ctx, x, w, h, c, n);                             // [t, c, h, w]
@@ -1393,7 +1392,7 @@ namespace WAN {
             k      = norm_k->forward(ctx, k);
             auto v = v_proj->forward(ctx, context);  // [N, n_context, dim]
 
-            x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, num_heads, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
+            x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, num_heads, nullptr, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
 
             x = o_proj->forward(ctx, x);  // [N, n_token, dim]
             return x;
@@ -1442,11 +1441,8 @@ namespace WAN {
             int64_t dim             = x->ne[0];
             int64_t context_txt_len = context->ne[1] - context_img_len;
 
-            context          = ggml_ext_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, context, 0, 2, 1, 3));  // [context_img_len + context_txt_len, N, dim]
+            auto context_img = ggml_view_3d(ctx->ggml_ctx, context, dim, context_img_len, N, context->nb[1], context->nb[2], 0);                                 // [N, context_img_len, dim]
-            auto context_img = ggml_view_3d(ctx->ggml_ctx, context, dim, N, context_img_len, context->nb[1], context->nb[2], 0);
+            auto context_txt = ggml_view_3d(ctx->ggml_ctx, context, dim, context_txt_len, N, context->nb[1], context->nb[2], context_img_len * context->nb[1]);  // [N, context_txt_len, dim]
-            auto context_txt = ggml_view_3d(ctx->ggml_ctx, context, dim, N, context_txt_len, context->nb[1], context->nb[2], context_img_len * context->nb[2]);
-            context_img      = ggml_ext_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, context_img, 0, 2, 1, 3));  // [N, context_img_len, dim]
-            context_txt      = ggml_ext_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, context_txt, 0, 2, 1, 3));  // [N, context_txt_len, dim]
 
             auto q = q_proj->forward(ctx, x);
             q      = norm_q->forward(ctx, q);
@@ -1458,8 +1454,8 @@ namespace WAN {
             k_img      = norm_k_img->forward(ctx, k_img);
             auto v_img = v_img_proj->forward(ctx, context_img);  // [N, context_img_len, dim]
 
-            auto img_x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k_img, v_img, num_heads, nullptr, false, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
+            auto img_x = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k_img, v_img, num_heads, nullptr, false, ctx->flash_attn_enabled);  // [N, n_token, dim]
-            x          = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, num_heads, nullptr, false, false, ctx->flash_attn_enabled);          // [N, n_token, dim]
+            x          = ggml_ext_attention_ext(ctx->ggml_ctx, ctx->backend, q, k, v, num_heads, nullptr, false, ctx->flash_attn_enabled);          // [N, n_token, dim]
 
             x = ggml_add(ctx->ggml_ctx, x, img_x);
 
@@ -1576,7 +1572,7 @@ namespace WAN {
             y = modulate_add(ctx->ggml_ctx, y, es[3]);
 
             y = ffn_0->forward(ctx, y);
-            y = ggml_gelu_inplace(ctx->ggml_ctx, y);
+            y = ggml_ext_gelu(ctx->ggml_ctx, y, true);
             y = ffn_2->forward(ctx, y);
 
             x = ggml_add(ctx->ggml_ctx, x, modulate_mul(ctx->ggml_ctx, y, es[5]));
@@ -1723,7 +1719,7 @@ namespace WAN {
 
             auto x = proj_0->forward(ctx, image_embeds);
             x      = proj_1->forward(ctx, x);
-            x      = ggml_gelu_inplace(ctx->ggml_ctx, x);
+            x      = ggml_ext_gelu(ctx->ggml_ctx, x, true);
             x      = proj_3->forward(ctx, x);
             x      = proj_4->forward(ctx, x);
 
@@ -1910,7 +1906,7 @@ namespace WAN {
             e0      = ggml_reshape_4d(ctx->ggml_ctx, e0, e0->ne[0] / 6, 6, e0->ne[1], e0->ne[2]);  //  [N, 6, dim] or [N, T, 6, dim]
 
             context = text_embedding_0->forward(ctx, context);
-            context = ggml_gelu(ctx->ggml_ctx, context);
+            context = ggml_ext_gelu(ctx->ggml_ctx, context);
             context = text_embedding_2->forward(ctx, context);  // [N, context_txt_len, dim]
 
             int64_t context_img_len = 0;
@@ -1949,7 +1945,7 @@ namespace WAN {
                     auto result = vace_block->forward(ctx, c, x_orig, e0, pe, context, context_img_len);
                     auto c_skip = result.first;
                     c           = result.second;
-                    c_skip      = ggml_scale(ctx->ggml_ctx, c_skip, vace_strength);
+                    c_skip      = ggml_ext_scale(ctx->ggml_ctx, c_skip, vace_strength);
                     x           = ggml_add(ctx->ggml_ctx, x, c_skip);
                 }
             }

src/z_image.hpp

@@ -54,15 +54,37 @@ namespace ZImage {
 
             auto qkv = qkv_proj->forward(ctx, x);                                                                            // [N, n_token, (num_heads + num_kv_heads*2)*head_dim]
             qkv      = ggml_reshape_4d(ctx->ggml_ctx, qkv, head_dim, num_heads + num_kv_heads * 2, qkv->ne[1], qkv->ne[2]);  // [N, n_token, num_heads + num_kv_heads*2, head_dim]
-            qkv      = ggml_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, qkv, 0, 2, 3, 1));                     // [num_heads + num_kv_heads*2, N, n_token, head_dim]
 
-            auto q = ggml_view_4d(ctx->ggml_ctx, qkv, qkv->ne[0], qkv->ne[1], qkv->ne[2], num_heads, qkv->nb[1], qkv->nb[2], qkv->nb[3], 0);                                           // [num_heads, N, n_token, head_dim]
+            auto q = ggml_view_4d(ctx->ggml_ctx,
-            auto k = ggml_view_4d(ctx->ggml_ctx, qkv, qkv->ne[0], qkv->ne[1], qkv->ne[2], num_kv_heads, qkv->nb[1], qkv->nb[2], qkv->nb[3], qkv->nb[3] * num_heads);                   // [num_kv_heads, N, n_token, head_dim]
+                                  qkv,
-            auto v = ggml_view_4d(ctx->ggml_ctx, qkv, qkv->ne[0], qkv->ne[1], qkv->ne[2], num_kv_heads, qkv->nb[1], qkv->nb[2], qkv->nb[3], qkv->nb[3] * (num_heads + num_kv_heads));  // [num_kv_heads, N, n_token, head_dim]
+                                  qkv->ne[0],
-
+                                  num_heads,
-            q = ggml_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, q, 0, 3, 1, 2));  // [N, n_token, num_heads, head_dim]
+                                  qkv->ne[2],
-            k = ggml_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, k, 0, 3, 1, 2));  // [N, n_token, num_kv_heads, head_dim]
+                                  qkv->ne[3],
-            v = ggml_cont(ctx->ggml_ctx, ggml_ext_torch_permute(ctx->ggml_ctx, v, 0, 3, 1, 2));  // [N, n_token, num_kv_heads, head_dim]
+                                  qkv->nb[1],
+                                  qkv->nb[2],
+                                  qkv->nb[3],
+                                  0);  // [N, n_token, num_heads, head_dim]
+            auto k = ggml_view_4d(ctx->ggml_ctx,
+                                  qkv,
+                                  qkv->ne[0],
+                                  num_kv_heads,
+                                  qkv->ne[2],
+                                  qkv->ne[3],
+                                  qkv->nb[1],
+                                  qkv->nb[2],
+                                  qkv->nb[3],
+                                  num_heads * qkv->nb[1]);  // [N, n_token, num_kv_heads, head_dim]
+            auto v = ggml_view_4d(ctx->ggml_ctx,
+                                  qkv,
+                                  qkv->ne[0],
+                                  num_kv_heads,
+                                  qkv->ne[2],
+                                  qkv->ne[3],
+                                  qkv->nb[1],
+                                  qkv->nb[2],
+                                  qkv->nb[3],
+                                  (num_heads + num_kv_heads) * qkv->nb[1]);  // [N, n_token, num_kv_heads, head_dim]
 
             if (qk_norm) {
                 auto q_norm = std::dynamic_pointer_cast<RMSNorm>(blocks["q_norm"]);
@@ -324,69 +346,6 @@ namespace ZImage {
             blocks["final_layer"] = std::make_shared<FinalLayer>(z_image_params.hidden_size, z_image_params.patch_size, z_image_params.out_channels);
         }
 
-        struct ggml_tensor* pad_to_patch_size(GGMLRunnerContext* ctx,
-                                              struct ggml_tensor* x) {
-            int64_t W = x->ne[0];
-            int64_t H = x->ne[1];
-
-            int pad_h = (z_image_params.patch_size - H % z_image_params.patch_size) % z_image_params.patch_size;
-            int pad_w = (z_image_params.patch_size - W % z_image_params.patch_size) % z_image_params.patch_size;
-            x         = ggml_ext_pad(ctx->ggml_ctx, x, pad_w, pad_h, 0, 0, ctx->circular_x_enabled, ctx->circular_y_enabled);
-            return x;
-        }
-
-        struct ggml_tensor* patchify(struct ggml_context* ctx,
-                                     struct ggml_tensor* x) {
-            // x: [N, C, H, W]
-            // return: [N, h*w, patch_size*patch_size*C]
-            int64_t N = x->ne[3];
-            int64_t C = x->ne[2];
-            int64_t H = x->ne[1];
-            int64_t W = x->ne[0];
-            int64_t p = z_image_params.patch_size;
-            int64_t h = H / z_image_params.patch_size;
-            int64_t w = W / z_image_params.patch_size;
-
-            GGML_ASSERT(h * p == H && w * p == W);
-
-            x = ggml_reshape_4d(ctx, x, p, w, p, h * C * N);                 // [N*C*h, p, w, p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));            // [N*C*h, w, p, p]
-            x = ggml_reshape_4d(ctx, x, p * p, w * h, C, N);                 // [N, C, h*w, p*p]
-            x = ggml_cont(ctx, ggml_ext_torch_permute(ctx, x, 2, 0, 1, 3));  // [N, h*w, C, p*p]
-            x = ggml_reshape_3d(ctx, x, C * p * p, w * h, N);                // [N, h*w, p*p*C]
-            return x;
-        }
-
-        struct ggml_tensor* process_img(GGMLRunnerContext* ctx,
-                                        struct ggml_tensor* x) {
-            x = pad_to_patch_size(ctx, x);
-            x = patchify(ctx->ggml_ctx, x);
-            return x;
-        }
-
-        struct ggml_tensor* unpatchify(struct ggml_context* ctx,
-                                       struct ggml_tensor* x,
-                                       int64_t h,
-                                       int64_t w) {
-            // x: [N, h*w, patch_size*patch_size*C]
-            // return: [N, C, H, W]
-            int64_t N = x->ne[2];
-            int64_t C = x->ne[0] / z_image_params.patch_size / z_image_params.patch_size;
-            int64_t H = h * z_image_params.patch_size;
-            int64_t W = w * z_image_params.patch_size;
-            int64_t p = z_image_params.patch_size;
-
-            GGML_ASSERT(C * p * p == x->ne[0]);
-
-            x = ggml_reshape_4d(ctx, x, C, p * p, w * h, N);                 // [N, h*w, p*p, C]
-            x = ggml_cont(ctx, ggml_ext_torch_permute(ctx, x, 1, 2, 0, 3));  // [N, C, h*w, p*p]
-            x = ggml_reshape_4d(ctx, x, p, p, w, h * C * N);                 // [N*C*h, w, p, p]
-            x = ggml_cont(ctx, ggml_permute(ctx, x, 0, 2, 1, 3));            // [N*C*h, p, w, p]
-            x = ggml_reshape_4d(ctx, x, W, H, C, N);                         // [N, C, h*p, w*p]
-
-            return x;
-        }
-
         struct ggml_tensor* forward_core(GGMLRunnerContext* ctx,
                                          struct ggml_tensor* x,
                                          struct ggml_tensor* timestep,
@@ -473,29 +432,24 @@ namespace ZImage {
             int64_t C = x->ne[2];
             int64_t N = x->ne[3];
 
-            auto img             = process_img(ctx, x);
+            int patch_size = z_image_params.patch_size;
+
+            auto img             = DiT::pad_and_patchify(ctx, x, patch_size, patch_size, false);
             uint64_t n_img_token = img->ne[1];
 
             if (ref_latents.size() > 0) {
                 for (ggml_tensor* ref : ref_latents) {
-                    ref = process_img(ctx, ref);
+                    ref = DiT::pad_and_patchify(ctx, ref, patch_size, patch_size, false);
                     img = ggml_concat(ctx->ggml_ctx, img, ref, 1);
                 }
             }
 
-            int64_t h_len = ((H + (z_image_params.patch_size / 2)) / z_image_params.patch_size);
-            int64_t w_len = ((W + (z_image_params.patch_size / 2)) / z_image_params.patch_size);
-
             auto out = forward_core(ctx, img, timestep, context, pe);
 
             out = ggml_ext_slice(ctx->ggml_ctx, out, 1, 0, n_img_token);                              // [N, n_img_token, ph*pw*C]
-            out = unpatchify(ctx->ggml_ctx, out, h_len, w_len);           // [N, C, H + pad_h, W + pad_w]
+            out = DiT::unpatchify_and_crop(ctx->ggml_ctx, out, H, W, patch_size, patch_size, false);  // [N, C, H, W]
 
-            // slice
+            out = ggml_ext_scale(ctx->ggml_ctx, out, -1.f);
-            out = ggml_ext_slice(ctx->ggml_ctx, out, 1, 0, H);  // [N, C, H, W + pad_w]
-            out = ggml_ext_slice(ctx->ggml_ctx, out, 0, 0, W);  // [N, C, H, W]
-
-            out = ggml_scale(ctx->ggml_ctx, out, -1.f);
 
             return out;
         }