DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-12 11:56:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

chore: fotmat code

Browse Source
This commit is contained in:
leejet committed 2026-11-01 13:40:36 +08:00
1 parent 6edeac1b20
commit 5531c8878a
24 files changed
+3052 -3379
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/core/cuda/sage_attention.cu
+3 -3
View File
@@ -287,9 +287,9 @@ namespace sd_cuda {
constexpr size_t shared_bytes = (128 + 64 + 64 * 2) * D;
auto kernel = sageattention::qk_int_sv_f16_attn_kernel<128, 64, 32, 64, D,
sageattention::DataType::kInt8, sageattention::QuantGranularity::kPerWarp,
sageattention::QuantGranularity::kPerWarp, float, false, half,
sageattention::ComputeUnit::kTensorCore, sageattention::MaskMode::kNone, false, false>;
sageattention::DataType::kInt8, sageattention::QuantGranularity::kPerWarp,
sageattention::QuantGranularity::kPerWarp, float, false, half,
sageattention::ComputeUnit::kTensorCore, sageattention::MaskMode::kNone, false, false>;
kernel<<<dim3(GGML_PAD(lq, 128) / 128, hq, batch), dim3(32, 4), shared_bytes, stream>>>(
qi, ki, static_cast<half*>(v->data), out, nullptr, qs, ks, nullptr,
lq, lk, hq / hk,