]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: ba58993)
metal : fix `ggml_metal_supports_op` for BF16 (#8021)
authorMichael de Gans <redacted>
Thu, 20 Jun 2024 05:32:01 +0000 (22:32 -0700)
committerGitHub <redacted>
Thu, 20 Jun 2024 05:32:01 +0000 (08:32 +0300)
commit2075a66a96cc1b04eabec7cf4b3051193d6f719e
tree4b4e30b099940c6aae4321a6394c2f15ab237bc3tree
parentba5899315283eb1df3902363daf79bdc5eefe426commit | diff
metal : fix `ggml_metal_supports_op` for BF16 (#8021)

Currently the Metal backend does not support BF16. `ggml_metal_supports_op` was returning true in these cases, leading to a crash with models converted with `--leave-output-tensor`. This commit checks if the first few sources types are BF16 and returns false if that's the case.
ggml-metal.m diff | blob | history
Packaging of ggml-org/llama.cpp