PR #2968 Add deepseekv3

Add deepseekv3 #2968

Narsil merged 11 commits into main from add_deepseekv3_2

Add fp8 support moe models

f190bc1d

Small modifications.

ee9178fb

Moe kernels 0.8.1

c8519aa7

Upgrade to 0.8.1

351f3c6e

Narsil force pushed from 1852d8ac to 351f3c6e 1 year ago

Fixing moe import.

75398810

Black.

9376066b

Narsil marked this pull request as ready for review 1 year ago

mht-sharma dismissed these changes on 2025-01-30

Apply suggestions from code review

f56e24b3

Narsil dismissed their stale review via f56e24b3 1 year ago

Narsil changed the title ~~[WIP] Add deepseekv3 2~~ Add deepseekv3 1 year ago

danieldk commented on 2025-01-30

Fixing Mixtral + Nits.

51bc8a4e

Put link to ref.

c174142f

Fix other call locations.

d5b2c25d

Scoring func `softmax` is the only one that works.

2c90b057

danieldk approved these changes on 2025-01-30

Narsil merged cb747b33 into main 1 year ago

Narsil deleted the add_deepseekv3_2 branch 1 year ago

Reviewers

danieldk

mht-sharma

Assignees

No one assigned

Labels

None yet

Milestone

No milestone