]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
cuda : add RDNA4-specific MMVQ parameter table for bs=1 decode (llama/19478)
authorPikaPikachu <redacted>
Sun, 15 Mar 2026 07:33:39 +0000 (15:33 +0800)
committerGeorgi Gerganov <redacted>
Sun, 15 Mar 2026 19:50:13 +0000 (21:50 +0200)
commit54042a3a28ac5d3910a8d76ca95fa7bddf5d926f
treecbc8dcad717532f1e93af9d2f9fe85fed43ad3d3
parentd596f17e771aa28bafbf90e7fa262bbf3a2e3d16
cuda : add RDNA4-specific MMVQ parameter table for bs=1 decode (llama/19478)

* mmvq: add RDNA3/RDNA4-specific parameter table (nwarps=8, rows=1)

* mmvq: add dedicated RDNA3 parameter table

* mmvq: exclude RDNA3.5 (gfx1150/1151) from RDNA3 table
src/ggml-cuda/mmvq.cu
src/ggml-cuda/vendors/hip.h