]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
overview / pkg / ggml / sources / ggml / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: a2bd034)
ggml-hexagon: flash-attn opt (llama/19025)
authornullname <redacted>
Sat, 24 Jan 2026 06:02:07 +0000 (14:02 +0800)
committerGeorgi Gerganov <redacted>
Fri, 30 Jan 2026 11:49:29 +0000 (13:49 +0200)
commit88b0e186b689fd7e95581fd29a81b969b2a23dae
tree54edbf6a8b3891a9ff552cae8f7fa73cc38e2fb2tree
parenta2bd034959543b67051d4ab9d50e15f55eabe011commit | diff
ggml-hexagon: flash-attn opt (llama/19025)

* optimize flash attention kernel by improving score computation and online softmax update

* wip

* Refactor online softmax update in flash attention kernel for improved performance

* Optimize flash attention kernel by replacing float array with HVX_Vector for score computation

* wip
src/ggml-hexagon/htp/flash-attn-ops.c diff | blob | history
Packaging of ggml-org/ggml