llama.cpp
ggml-webgpu: enable FLASH_ATTN_EXT on browser without subgroup matrix
#22199
Merged
Merged
ggml-webgpu: enable FLASH_ATTN_EXT on browser without subgroup matrix #22199
reeselevine
merged
14 commits into
ggml-org:master
from
ArberSephirotheca:browser
ArberSephirotheca
changed the title 
ggml-webgpu: add tile flash attention fallback
47e4de31
ggml-webgpu: add new fields and discard usage of mnk for tile version
54974e3f
ggml-webgpu: modify the vec path to discard the mnk parameter
39d7f280
ggml-webgpu: enable flash attention vec and tile version for broswer
e0f8d890
ggml-webgpu: stagging KV for flash attention tile version
6d210998
formatting
a144c47d
turn on subgroup uniformity check
2a43134f
remove Q_TILE as it is always 1 for vec path
28cf9535
make row_max and exp_sum to local register
1fed0ee8
make different bindings with same underlying buffer to have the same …
e60c49bd
move path selection into the shader library and have the host consume…
35add3da
turn off skip_validation and address buffer overlapping when nwg==1
ed8bb6cf
formatting
4c0e94a8
merge binding when kv overlap
a0a3f856
reeselevine
approved these changes
on 2026-04-24
CISC
approved these changes
on 2026-04-24
Reviewers
Assignees
No one assigned
Labels
ggml
WebGPU
Milestone
No milestone
Login to write a write a comment.
Login via GitHub