]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
overview / pkg / ggml / sources / ggml / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 90fac5e)
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>
Tue, 21 Oct 2025 15:14:33 +0000 (18:14 +0300)
commit1fe7675fa66ca8c6930f93b54c3b40c8d964276b
treec47b5016bccd8a586cc4d7160c8e6fe2b8465aa9tree
parent90fac5e0309f3d88890175a1283b64f4042b852bcommit | diff
CUDA: use registers instead of smem in topk-moe (llama/16647)

Uses the technique used in the vulkan PR #16641. Neat trick!
src/ggml-cuda/topk-moe.cu diff | blob | history
Packaging of ggml-org/ggml