PR #30640 llama: create new fused op for QSA mask

llama: create new fused op for QSA mask #30640

am17an merged 4 commits into master from aman/cuda-kpool-mask

am17an requested a review from

ggerganov 14 days ago

am17an requested a review from

CISC 14 days ago

am17an requested a review 14 days ago

am17an requested a review from

JohannesGaessler 14 days ago

am17an changed the title ~~llama: create new fused op for QSA scoring~~ llama: create new fused op for QSA mask 14 days ago

github-actions added model

github-actions added testing

github-actions added ggml

github-actions added CUDA

am17an force pushed from ee0d2cc9 to 8a862123 14 days ago

llama: create new fused op for QSA scoring

4ac70767

enroll glm5.3-next too

6459705e

tests : initialize the remaining tensors in test_kpool_mask

2961d5e5

am17an force pushed from 8a862123 to 2961d5e5 9 days ago

ggerganov assigned

ggerganov 9 days ago

ServeurpersoCom requested a review from

ServeurpersoCom 9 days ago

ServeurpersoCom approved these changes on 2026-10-21

ggerganov commented on 2026-10-21

ggml: generalize the k-pool mask op into ggml_top_k_mask

55672a51

am17an force pushed from 55672a51 to 2961d5e5 9 days ago

ggerganov approved these changes on 2026-10-21

am17an merged d454a204 into master 9 days ago

Reviewers

ggerganov

ServeurpersoCom

CISC

JohannesGaessler

Assignees

ggerganov

Labels

model testing ggml CUDA

Milestone

No milestone