Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add flash attention MMA / Tiles to support MiMo-V2.5 (#22812)

Browse Source 
* mimo-v2.5: add flash attention mma/tiles for for d_kq=192 d_v=128

* mimo-v2.5: follow (256, 256) fattn templates

* mimo-v2.5: cleanup comments

* mimo-v2.5: further comment cleanup

* mimo-v2.5: address PR feedback
fix GQA handling
check for other dangling 320/576 carveouts and mirror them for 192
Add to backend ops test so new paths are covered
This commit is contained in:
AesSedai
2026-05-09 11:28:29 +08:00
committed by GitHub
parent 66001722aa
commit 046e284437
14 changed files with 105 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/fattn-mma-f16.cuh
+9
View File
@@ -61,6 +61,11 @@ static constexpr __host__ __device__ fattn_mma_config ggml_cuda_fattn_mma_get_co
GGML_CUDA_FATTN_MMA_CONFIG_CASE(128, 128, 32, 128, 2, 64, 64, 64, 64, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(128, 128, 64, 128, 2, 64, 64, 64, 64, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(192, 128, 8, 64, 4, 64, 96, 64, 64, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(192, 128, 16, 64, 4, 32, 96, 64, 64, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(192, 128, 32, 128, 2, 32, 96, 64, 64, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(192, 128, 64, 128, 2, 32, 96, 64, 64, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(256, 256, 8, 64, 4, 64, 128, 128, 128, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(256, 256, 16, 64, 4, 32, 128, 128, 128, 2, true);
GGML_CUDA_FATTN_MMA_CONFIG_CASE(256, 256, 32, 128, 2, 32, 128, 128, 128, 2, true);
@@ -1561,6 +1566,10 @@ static __global__ void flash_attn_ext_f16(
NO_DEVICE_CODE;
return;
}
if (DKQ == 192 && ncols2 != 8 && ncols2 != 16) {
NO_DEVICE_CODE;
return;
}
#ifdef VOLTA_MMA_AVAILABLE
if (ncols1*ncols2 < 32) {
NO_DEVICE_CODE;