llama.cpp
SYCL: add BF16 to DMMV kernel path (~4x tg speedup on Intel Arc)
#21580
Merged

SYCL: add BF16 to DMMV kernel path (~4x tg speedup on Intel Arc) #21580

ggerganov merged 2 commits into ggml-org:master from PMZFX:sycl-bf16-dmmv
PMZFX
PMZFX SYCL: add BF16 to DMMV kernel path for ~4x token generation speedup
b86ed2ea
PMZFX PMZFX requested a review 92 days ago
github-actions github-actions added ggml
github-actions github-actions added SYCL
NeoZhangJianyu
simonlui
simonlui commented on 2026-05-02
arthw
arthw commented on 2026-05-08
arthw
PMZFX SYCL: fix BF16 DMMV out-of-bounds when ncols % 64 != 0
17052fa7
PMZFX
arthw
arthw approved these changes on 2026-05-15
arthw arthw added merge ready
ggerganov ggerganov merged 8cc67efc into master 48 days ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone