]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
cuda : add f32 to bf16 copy op (llama/12806)
authorSigbjørn Skjæret <redacted>
Tue, 8 Apr 2025 21:21:31 +0000 (23:21 +0200)
committerGeorgi Gerganov <redacted>
Thu, 10 Apr 2025 20:58:06 +0000 (23:58 +0300)
commitd206e725693e8b06f7e5af33b6fe2bb5aa25e1ef
treee5d63abc1e8e6609a1c0f756df87f7ee25ace544
parent637503b4eb061004850fe8e7c4e6962ae1d66d94
cuda : add f32 to bf16 copy op (llama/12806)

This allows BF16 KV-cache on CUDA.
src/ggml-cuda/cpy.cu
src/ggml-cuda/ggml-cuda.cu