PR #13269 Model: Granite MoE shared

Model: Granite MoE shared #13269

CISC merged 11 commits into ggml-org:master from gabe-l-hart:GraniteMoEShared

github-actions added python

ngxson commented on 2025-05-02

feat: Add GGUF conversion for granitemoeshared

a6529ccd

feat: hparam and arch plumbing for granitemoeshared

731c5fc4

fix: Split MoE fused tensors for shared experts in conversion

c5d897ed

feat: First WIP cut at model arch in cpp

054059ea

fix: Cleaner (maybe more correct?) splitting for gate/up

5a98b485

fix: Fix the input to the shared experts

9763c9a2

fix: Avoid architecture-specific checks for Granite MoE Shared

52d2ed6e

gabe-l-hart force pushed from 97de56d8 to 52d2ed6e 234 days ago

refactor: Split granite architectures out of llm_build_llama

44469949

fix: Fix compiler warning about uninitialized inp_pos

3d792146

gabe-l-hart force pushed to 3d792146 234 days ago

ggerganov approved these changes on 2025-05-12

fix: Consoladate GraniteMoEShared into GraniteMoE for conversion

2aed91c3

fix: Consolidate GraniteMoEShared into GraniteMoE on the c++ side

33008e8c

CISC merged d590cd4c into master 231 days ago

gabe-l-hart deleted the GraniteMoEShared branch 231 days ago

ggerganov commented on 2025-05-14

Reviewers

ggerganov

ngxson

Assignees

No one assigned

Labels

python

Milestone

No milestone