]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
CUDA: use fastdiv + ggml_cuda_mad for mmvf (llama/16557)
authorAman Gupta <redacted>
Tue, 14 Oct 2025 11:16:21 +0000 (19:16 +0800)
committerGeorgi Gerganov <redacted>
Wed, 15 Oct 2025 06:29:17 +0000 (09:29 +0300)
commitf2075667fa872b95b8afa3517f938432ffb488ba
tree409ff6554209909c17862e65f32545bb2dd40a4b
parentb4c5c6f71fec4c04a1805df73588f52d316f31ce
CUDA: use fastdiv + ggml_cuda_mad for mmvf (llama/16557)

* CUDA: use fastdiv + ggml_cuda_mad for mmvf

* use bf16 directly + fix formatting

* Add exception for HIP code
ggml/src/ggml-cuda/mmvf.cu