]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
overview / pkg / ggml / sources / ggml / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 2327bda)
metal : single allocation of encode_async block (llama/9747)
authorPaul Tsochantaris <redacted>
Mon, 7 Oct 2024 12:26:31 +0000 (13:26 +0100)
committerGeorgi Gerganov <redacted>
Wed, 16 Oct 2024 08:28:38 +0000 (11:28 +0300)
commit89db9766ae28b2dc712296e41226ffe6bdb623a7
tree4dc9995d048ec54979b767b9f4d5f57adfb8a5f9tree
parent2327bda7a55ac6b72614ac5ebd5c5a5e02553b9bcommit | diff
metal : single allocation of encode_async block (llama/9747)

* Single allocation of encode_async block with non-ARC capture in ggml-metal.m

* Moving Block_release to the deallocation code

* Release encode block when re-setting encoding buffer count if needed

* Update ggml/src/ggml-metal.m

---------

Co-authored-by: Georgi Gerganov <redacted>
src/ggml-metal.m diff | blob | history
Packaging of ggml-org/ggml