]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
CUDA: use registers instead of smem in topk-moe (llama/16647)
authorAman Gupta <redacted>
Sat, 18 Oct 2025 09:52:53 +0000 (17:52 +0800)
committerGeorgi Gerganov <redacted>
Wed, 22 Oct 2025 09:58:11 +0000 (12:58 +0300)
commit08345f15ece9bdc528770596c08b48144082e933
treec27b332a9d7cf561fd87023f3ae5bdcade0d71b8
parent8ffdf4bd963bfe4437f35620d884884055a68f64
CUDA: use registers instead of smem in topk-moe (llama/16647)

Uses the technique used in the vulkan PR #16641. Neat trick!
ggml/src/ggml-cuda/topk-moe.cu