PR #19012 ggml-cpu: Use tiled FA for prompt-processing

ggml-cpu: Use tiled FA for prompt-processing #19012

am17an merged 6 commits into ggml-org:master from am17an:tile-fa-cpu

am17an requested a review from

ggerganov 155 days ago

github-actions added ggml

am17an force pushed from df10660b to 97afbcbb 155 days ago

am17an force pushed from 97afbcbb to 41a07185 155 days ago

ggml-cpu: Use tiled FA for prompt-processing

2f09b2d3

am17an force pushed from 41a07185 to 2f09b2d3 154 days ago

ggerganov commented on 2026-01-23

fix out of bounds for mask

e30395e5

skip rows where there are all masks

693935d9

ggerganov commented on 2026-01-24

skip tile if mask is inf

d898d43a

ggerganov commented on 2026-01-24

store mask in worksize

dc30629d

am17an force pushed from c1dbc374 to dc30629d 152 days ago

ggerganov approved these changes on 2026-01-25

check inf tile earlier

17f7db50

am17an merged bcb43163 into master 151 days ago

am17an deleted the tile-fa-cpu branch 151 days ago

Reviewers

ggerganov

Assignees

No one assigned

Labels

ggml

Milestone

No milestone