]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: b38a16d)
ggml : re-enable BLAS for CPU when src0 != F32 + remove redundant full offload checks...
authorGeorgi Gerganov <redacted>
Tue, 28 Nov 2023 08:32:03 +0000 (10:32 +0200)
committerGitHub <redacted>
Tue, 28 Nov 2023 08:32:03 +0000 (10:32 +0200)
commit8406b0924bf323f37d536dee8b8165c1f3d9d11d
tree23b753a4e289a89c9e511760b8d2e6583fda9768tree
parentb38a16dfcff88d547f78f52d1bea31b84a05aff7commit | diff
ggml : re-enable BLAS for CPU when src0 != F32 + remove redundant full offload checks in llama.cpp (#4240)

* ggml : use blas even if src0 is not F32

* llama : use n_threads_batch only when n_tokens >= 32

ggml-ci

* llama : revert n_threads_batch logic

ggml-ci
ggml.c diff | blob | history
llama.cpp diff | blob | history
Packaging of ggml-org/llama.cpp