refactor: split model loader from model definitions (#1619)

src/model/adapter/lora.hpp

@@ -3,6 +3,7 @@
 
 #include <mutex>
 #include "core/ggml_extend.hpp"
+#include "model_loader.h"
 
 #define LORA_GRAPH_BASE_SIZE 10240
 

src/model/adapter/pmid.hpp

@@ -6,6 +6,7 @@
 #include "model/adapter/lora.hpp"
 #include "model/common/block.hpp"
 #include "model/te/clip.hpp"
+#include "model_loader.h"
 
 struct FuseBlock : public GGMLBlock {
     // network hparams

src/model/diffusion/control.hpp

@@ -1,8 +1,8 @@
 #ifndef __SD_MODEL_DIFFUSION_CONTROL_HPP__
 #define __SD_MODEL_DIFFUSION_CONTROL_HPP__
 
-#include "model.h"
 #include "model/common/block.hpp"
+#include "model_loader.h"
 
 #define CONTROL_NET_GRAPH_SIZE 1536
 

src/model/diffusion/flux.hpp

@@ -4,10 +4,10 @@
 #include <memory>
 #include <vector>
 
-#include "model.h"
 #include "model/common/rope.hpp"
 #include "model/diffusion/dit.hpp"
 #include "model/diffusion/model.hpp"
+#include "model_loader.h"
 
 #define FLUX_GRAPH_SIZE 10240
 

src/model/diffusion/ltxv.hpp

@@ -13,6 +13,7 @@
 #include "model/common/rope.hpp"
 #include "model/diffusion/flux.hpp"
 #include "model/diffusion/model.hpp"
+#include "model_loader.h"
 
 namespace LTXV {
 

src/model/diffusion/mmdit.hpp

@@ -7,9 +7,9 @@
 #include <vector>
 
 #include "core/ggml_extend.hpp"
-#include "model.h"
 #include "model/common/block.hpp"
 #include "model/diffusion/model.hpp"
+#include "model_loader.h"
 
 #define MMDIT_GRAPH_SIZE 10240
 

src/model/diffusion/qwen_image.hpp

@@ -6,6 +6,7 @@
 #include "model/common/block.hpp"
 #include "model/diffusion/flux.hpp"
 #include "model/diffusion/model.hpp"
+#include "model_loader.h"
 
 namespace Qwen {
     constexpr int QWEN_IMAGE_GRAPH_SIZE = 20480;

src/model/diffusion/wan.hpp

@@ -9,6 +9,7 @@
 #include "model/common/rope.hpp"
 #include "model/diffusion/flux.hpp"
 #include "model/diffusion/model.hpp"
+#include "model_loader.h"
 
 namespace WAN {
 

src/model/diffusion/z_image.hpp

@@ -7,6 +7,7 @@
 #include "model/diffusion/flux.hpp"
 #include "model/diffusion/mmdit.hpp"
 #include "model/diffusion/model.hpp"
+#include "model_loader.h"
 
 // Ref: https://github.com/Alpha-VLLM/Lumina-Image-2.0/blob/main/model/model.py
 // Ref: https://github.com/huggingface/diffusers/pull/12703

src/model/te/llm.hpp

@@ -21,6 +21,7 @@
 #include "core/ggml_extend.hpp"
 #include "json.hpp"
 #include "model/common/rope.hpp"
+#include "model_loader.h"
 #include "tokenizers/bpe_tokenizer.h"
 #include "tokenizers/gemma_tokenizer.h"
 #include "tokenizers/gpt_oss_tokenizer.h"

src/model/te/t5.hpp

@@ -11,7 +11,7 @@
 #include <unordered_map>
 
 #include "core/ggml_extend.hpp"
-#include "model.h"
+#include "model_loader.h"
 #include "tokenizers/t5_unigram_tokenizer.h"
 
 struct T5Config {

src/model/upscaler/esrgan.hpp

@@ -2,7 +2,7 @@
 #define __SD_MODEL_UPSCALER_ESRGAN_HPP__
 
 #include "core/ggml_extend.hpp"
-#include "model.h"
+#include "model_loader.h"
 
 /*
     ===================================    ESRGAN  ===================================

src/model/upscaler/ltx_latent_upscaler.hpp

@@ -14,8 +14,8 @@
 #include "core/ggml_extend.hpp"
 #include "core/ggml_graph_cut.h"
 #include "core/util.h"
-#include "model.h"
 #include "model/diffusion/dit.hpp"
+#include "model_loader.h"
 
 namespace LTXVUpsampler {
     constexpr int LTX_UPSAMPLER_GRAPH_SIZE = 10240;

src/model/vae/ltx_audio_vae.hpp

@@ -8,6 +8,7 @@
 #include <vector>
 
 #include "core/ggml_extend.hpp"
+#include "model_loader.h"
 
 namespace LTXV {
 

src/model/vae/ltx_vae.hpp

@@ -12,6 +12,7 @@
 #include "model/diffusion/ltxv.hpp"
 #include "model/vae/vae.hpp"
 #include "model/vae/wan_vae.hpp"
+#include "model_loader.h"
 
 namespace LTXVAE {
 

src/model/vae/wan_vae.hpp

@@ -7,6 +7,7 @@
 
 #include "model/common/block.hpp"
 #include "model/vae/vae.hpp"
+#include "model_loader.h"
 
 namespace WAN {