]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
CUDA: fuse rope + set_rows (#16884)
authorAman Gupta <redacted>
Thu, 13 Nov 2025 00:50:01 +0000 (08:50 +0800)
committerGitHub <redacted>
Thu, 13 Nov 2025 00:50:01 +0000 (08:50 +0800)
commita90eb94ca9ec19f049a1c8e4958e71d9da777569
tree66443588892a66d57afeb7723268a445d3023bc0
parent07751f8d446e2d05d069e8d77d984dd64c1a5878
CUDA: fuse rope + set_rows (#16884)

* CUDA: add fused rope

* move k forward_expand up

* create helper function instead of re-using params

* make assert statement more in line with comment

* rope_norm: coalesced writes to global mem
ggml/src/ggml-cuda/ggml-cuda.cu
ggml/src/ggml-cuda/rope.cu
ggml/src/ggml-cuda/rope.cuh
src/llama-graph.cpp