CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-19 11:57:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: optimize FA for GQA + large batches (#12014)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-02-22 12:20:17 +01:00
committed by GitHub
parent 335eb04a91
commit 5fa07c2f93
32 changed files with 940 additions and 411 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda/fattn-vec-f16.cuh
View File

@@ -310,7 +310,7 @@ void ggml_cuda_flash_attn_ext_vec_f16_case_impl(ggml_backend_cuda_context & ctx,
constexpr bool need_f16_K = D != 128;
constexpr bool need_f16_V = D != 128 && D != 64;
constexpr size_t nbytes_shared = 0;
launch_fattn<D, cols_per_block, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, nbytes_shared, need_f16_K, need_f16_V);
launch_fattn<D, cols_per_block, 1, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, nbytes_shared, need_f16_K, need_f16_V);
}
template <int D, ggml_type type_K, ggml_type type_V>