llama.cpp
llama : scale moe down projection per expert token
#30861

Merged

llama : scale moe down projection per expert token #30861

ggerganov merged 4 commits into master from gg/amax-pool1d-meta

github-actions added ggml

ggerganov commented on 2026-10-20

ggerganov force pushed 15 days ago

ggerganov marked this pull request as ready for review 15 days ago

ggerganov requested a review from

CISC 15 days ago

ggerganov requested a review from

JohannesGaessler 15 days ago

llama : scale moe down projection per expert token

1918cd5a

cont : simplify comments

0c5b7cbb

ggerganov force pushed to 0c5b7cbb 15 days ago

llama : refactor amax src1 scaling result

4019bf9a

cont : add TODO + clean-up

6bd4567b

ggerganov added merge ready

ggerganov merged 8baabd53 into master 15 days ago

ggerganov deleted the gg/amax-pool1d-meta branch 15 days ago

JohannesGaessler commented on 2026-10-23

Reviewers

JohannesGaessler

CISC

Assignees

No one assigned

Labels

ggml merge ready

Milestone

No milestone