]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 87f4744)
gemma : perform per-layer projections in the first layer (#21612)
authorGeorgi Gerganov <redacted>
Wed, 8 Apr 2026 13:06:30 +0000 (16:06 +0300)
committerGitHub <redacted>
Wed, 8 Apr 2026 13:06:30 +0000 (16:06 +0300)
commit5764d7c6a64b57e3cfbc124451666407ab47ee26
treef40b6e63389cbfd7399ed4f575ad20755d171d1btree
parent87f4744a805f7837b95ce7ff94fc60eb7a3fbcc8commit | diff
gemma : perform per-layer projections in the first layer (#21612)

* gemma : reduce graph splits by keeping per-layer ops in the input layer

* gemma : put the per-layer proj in the first layer

* cont : move the projection before the layer loop
src/llama-arch.cpp diff | blob | history
src/llama-model.cpp diff | blob | history
src/llama-model.h diff | blob | history
src/models/gemma3n-iswa.cpp diff | blob | history
src/models/gemma4-iswa.cpp diff | blob | history
src/models/models.h diff | blob | history
Packaging of ggml-org/llama.cpp