]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: c50b34a)
opencl: route large q6_K lm_head to the flat GEMV (#26427)
authorHongqiang Wang <redacted>
Mon, 3 Aug 2026 14:36:19 +0000 (07:36 -0700)
committerGitHub <redacted>
Mon, 3 Aug 2026 14:36:19 +0000 (07:36 -0700)
commit39eab74a05d3e68ac822b6dc6cd78c90cb985c19
treebee33c4cd4693a862d1d5b9f1a948eaf0a271e3btree
parentc50b34a1e0330c4781a0846932033689ee1eaa85commit | diff
opencl: route large q6_K lm_head to the flat GEMV (#26427)

* add a direct size condition for `large` weights; the original
  dimension condition is insufficient -- q6_K lm_head for gemma-4 E2B
  has [1536, 262144], which is big enough to slowdown gemv_noshuffle but
  does not satisfy the dimension condition (ne0 >= 2048)
ggml/src/ggml-opencl/ggml-opencl.cpp diff | blob | history
Packaging of ggml-org/llama.cpp