{"entries":[{"name":"CMakeLists.txt","mode":"0100644","size":4291,"lastCommit":{"hash":"db9d8aa428012cc5593e18635d4c3c54095f5138","oid":"db9d8aa428012cc5593e18635d4c3c54095f5138","parents":["ccb87fa3ee1961ec915f77cb447706f471dca6a5"],"author":{"name":"Patrick Buckley","email":"eous@users.noreply.github.com","when":"2026-03-22T03:05:51-07:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2026-03-22T11:05:51+01:00"},"message":"ggml-cuda: native bf16 flash attention for vec kernel (#20525)\n\n* ggml-cuda: native bf16 flash attention for vec and tile kernels\n\nmma kernel still converts bf16 to fp16 before launch, native mma bf16 todo\n\n* ggml-cuda: address code owner review feedback\n\nreverted tile kernel changes to avoid larger refactor\n\n* fix ci failures on turing and hip\n\n* fix bf16 vec kernel compile on hip v_dot2 platforms\n\n* add comments\n\n---------\n\nCo-authored-by: Johannes Gäßler ","tree":"8c5e856b43f1b93ef769abba876839610f750f5b"}},{"name":"mudnn.cu","mode":"0100644","size":3794,"lastCommit":{"hash":"33983057d0f578aca74ba15eccc3de9c267a5ff6","oid":"33983057d0f578aca74ba15eccc3de9c267a5ff6","parents":["fb1cab201c6c4cd36731f100df74eece2f4706fa"],"author":{"name":"R0CKSTAR","email":"yeahdongcn@gmail.com","when":"2025-05-21T09:58:49+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-05-21T09:58:49+08:00"},"message":"musa: Upgrade MUSA SDK version to rc4.0.1 and use mudnn::Unary::IDENTITY op to accelerate D2D memory copy (#13647)\n\n* musa: fix build warning (unused parameter)\n\nSigned-off-by: Xiaodong Ye \n\n* musa: upgrade MUSA SDK version to rc4.0.1\n\nSigned-off-by: Xiaodong Ye \n\n* musa: use mudnn::Unary::IDENTITY op to accelerate D2D memory copy\n\nSigned-off-by: Xiaodong Ye \n\n* Update ggml/src/ggml-cuda/cpy.cu\n\nCo-authored-by: Johannes Gäßler \n\n* musa: remove MUDNN_CHECK_GEN and use CUDA_CHECK_GEN instead in MUDNN_CHECK\n\nSigned-off-by: Xiaodong Ye \n\n---------\n\nSigned-off-by: Xiaodong Ye \nCo-authored-by: Johannes Gäßler ","tree":"16ec7da4508db368e8aac529a3f5153c7d326824"}},{"name":"mudnn.cuh","mode":"0100644","size":333,"lastCommit":{"hash":"716301d1b03c31875ec3b24526c48c8b1bd0fd8c","oid":"716301d1b03c31875ec3b24526c48c8b1bd0fd8c","parents":["60ef23d6c14d325d83eae5752e5de39ad268e9b0"],"author":{"name":"R0CKSTAR","email":"yeahdongcn@gmail.com","when":"2025-06-26T12:11:59+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-06-26T12:11:59+08:00"},"message":"musa: enable fp16 mma (all) and cublas on qy2 (#13842)\n\n* musa: enable fp16 mma (all) and cublas on qy2\n\nSigned-off-by: Xiaodong Ye \n\n* Update ggml/src/ggml-cuda/ggml-cuda.cu\n\nCo-authored-by: Johannes Gäßler \n\n* Address review comments\n\nSigned-off-by: Xiaodong Ye \n\n* Address review comments\n\nSigned-off-by: Xiaodong Ye \n\n* musa: disable MUL_MAT_ID (q2_k × f32) due to precision issues\n\nSigned-off-by: Xiaodong Ye \n\n---------\n\nSigned-off-by: Xiaodong Ye \nCo-authored-by: Johannes Gäßler ","tree":"b5e0408499177b9c5d9ada0a175a37fe5e8d2078"}}],"lastCommit":{"hash":"db9d8aa428012cc5593e18635d4c3c54095f5138","oid":"db9d8aa428012cc5593e18635d4c3c54095f5138","parents":["ccb87fa3ee1961ec915f77cb447706f471dca6a5"],"author":{"name":"Patrick Buckley","email":"eous@users.noreply.github.com","when":"2026-03-22T03:05:51-07:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2026-03-22T11:05:51+01:00"},"message":"ggml-cuda: native bf16 flash attention for vec kernel (#20525)\n\n* ggml-cuda: native bf16 flash attention for vec and tile kernels\n\nmma kernel still converts bf16 to fp16 before launch, native mma bf16 todo\n\n* ggml-cuda: address code owner review feedback\n\nreverted tile kernel changes to avoid larger refactor\n\n* fix ci failures on turing and hip\n\n* fix bf16 vec kernel compile on hip v_dot2 platforms\n\n* add comments\n\n---------\n\nCo-authored-by: Johannes Gäßler ","tree":"8c5e856b43f1b93ef769abba876839610f750f5b"}}