llama.cpp
Extended SYCL oneDNN SDPA to non-FP16 KV caches (Q4_0–Q8_0 and FP32)
#25874

Merged

Extended SYCL oneDNN SDPA to non-FP16 KV caches (Q4_0–Q8_0 and FP32) #25874

arthw merged 2 commits into ggml-org:master from johnkarlhill:sycl-onednn-fa-quants

github-actions added documentation

github-actions added ggml

github-actions added SYCL

johnkarlhill force pushed from 0d01c9d9 to b0b76ae5 25 days ago

arthw commented on 2026-07-22

arthw approved these changes on 2026-07-27

johnkarlhill force pushed from 6e9257c1 to 2736decd 14 days ago

johnkarlhill marked this pull request as ready for review 14 days ago

johnkarlhill requested a review 14 days ago

arthw added merge ready

sycl: extend oneDNN SDPA to Q4_0-Q8_0 and F32 KV caches

ee36b4d3

docs: drop GGML_SYCL_FA_DEBUG from SYCL.md (not shipped in this PR)

d688c52a

johnkarlhill force pushed from 2736decd to d688c52a 13 days ago

CISC approved these changes on 2026-08-01

arthw merged 66fa168a into master 9 days ago

Reviewers

CISC

arthw

Assignees

No one assigned

Labels

documentation ggml merge ready SYCL

Milestone

No milestone