llama.cpp
ggml-webgpu: FlashAttention refactor + standardize quantization support
#23834

Merged

ggml-webgpu: FlashAttention refactor + standardize quantization support #23834

ggerganov merged 9 commits into ggml-org:master from reeselevine:flash_attn_refactor

Start work on flash_attn refactor

e1a0bb34

Refactor

8d61b5cd

Split k/v quantization

19f150a5

Refactor and abstract quantization logic for flash_attn and mul_mat

94a1fdde

Add quantization support to tile path

ddedf2fc

formatting

ffbd86f0

github-actions added ggml

github-actions added WebGPU

Merge remote-tracking branch 'origin/master' into flash_attn_refactor

2e1d2b5c

Merge remote-tracking branch 'upstream/master' into flash_attn_refactor

e0868e34

reeselevine marked this pull request as ready for review 111 days ago

reeselevine requested a review 111 days ago

yomaytk commented on 2026-06-02

Move to functions, add a check

cf66f0a3

reeselevine requested a review from

ggerganov 110 days ago

github-actions added testing

yomaytk approved these changes on 2026-06-03

reeselevine added merge ready

ggerganov merged e8c54893 into master 109 days ago

Reviewers

yomaytk

ggerganov

Assignees

No one assigned

Labels

testing ggml merge ready WebGPU

Milestone

No milestone