]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
HIP: use hipBLAS for dense prefill on gfx900, keep MMQ for MoE (llama/24588)
authorzduford <redacted>
Tue, 30 Jun 2026 09:51:38 +0000 (05:51 -0400)
committerGeorgi Gerganov <redacted>
Fri, 10 Jul 2026 10:06:42 +0000 (13:06 +0300)
commit2ea7a752afd5deca7e366dee18c7500b5173c69f
tree2ba90a4686f738edf8c732cc6234393cf08db210
parentb313fd3ce8c151a0ddc18c85cf40b76c1bec4958
HIP: use hipBLAS for dense prefill on gfx900, keep MMQ for MoE (llama/24588)

* HIP: keep MMQ for gfx900 MoE and Q8_0, use hipBLAS for dense K-quants

Assisted-by: GitHub Copilot CLI
* HIP: tighten conditional block to be explicitly for gfx900

* HIP: Further simplified gfx900 conditional block

* removed unnecessary comment
ggml/src/ggml-cuda/mmq.cu