llama.cpp f74e2fa for NVIDIA Volta (sm_70), CUDA 12.9 — released 2026-10-20 with the Q5_1 caveat stated
0b864c5 verified Download sm70-flash-attention.patch from Atlas3D/llama.cpp-sm70-cuda12.9: direct link, hf CLI and curl.
- Browser
- Download file 417 Bytes
-
https://huggingface.co/Atlas3D/llama.cpp-sm70-cuda12.9/resolve/main/sm70-flash-attention.patch
- Command line
-
hf download hf://Atlas3D/llama.cpp-sm70-cuda12.9/sm70-flash-attention.patch
-
curl -L -o sm70-flash-attention.patch https://huggingface.co/Atlas3D/llama.cpp-sm70-cuda12.9/resolve/main/sm70-flash-attention.patch
417 Bytes
| --- a/ggml/src/ggml-cuda/fattn.cu | |
| +++ b/ggml/src/ggml-cuda/fattn.cu | |
| } | |
| if constexpr (ncols2 <= 16) { | |
| - if (Q->ne[1] <= 16/ncols2) { | |
| + // Volta MMA kernels require at least 32 columns. | |
| + if (!volta_mma_available(cc) && Q->ne[1] <= 16/ncols2) { | |
| ggml_cuda_flash_attn_ext_mma_f16_case<DKQ, DV, 16/ncols2, ncols2>(ctx, dst); | |
| return; | |
| } | |