[model] feat: add MiniMax-M2 MoE bridge - #2602

Merged

yaoyu-33 merged 45 commits into

mainfrom

yuya/add-minimax-m2-bridge

Mar 23, 2026

Commits on Mar 6, 2026

[ckpt] refactor: Consolidate fused expert mappings and fix MTP inference
yaoyu-33
committed
ff3705b
renmove deprecated providers
yaoyu-33
committed
cf6391d
fix: resolve import errors from removed model providers
yaoyu-33
committed
755b07d
[model] feat: add MiniMax-M2 MoE bridge
yaoyu-33
committed
b2382d1
[model] feat: add full-dimension QK norm, FP8 dequantization, and multi-node support for MiniMax-M2
yaoyu-33
committed
40a9ca8
[model] fix: Add GLMExpertGateUpProjMapping alias and clean up MiniMax-M2 expert mappings
yaoyu-33
committed
a86dc26
[model] chore: clean up MiniMax-M2 example scripts
yaoyu-33
committed
f86b7ad

Commits on Mar 7, 2026

Commits on Mar 10, 2026

Commits on Mar 16, 2026

[model] chore: merge main into refactor-fused-expert-mappings

yaoyu-33
and
claude
committed
46da10c
Merge branch 'main' into yuya/add-minimax-m2-bridge
yaoyu-33
committed
c27268b
[model, ckpt] fix: fix GPT-OSS MXFP4 expert weight layout and GLM MTP inference

yaoyu-33
and
claude
committed
0e4d6ed

Commits on Mar 17, 2026

[ckpt] fix: Address PR review comments on fused expert mappings

yaoyu-33
and
claude
committed
efb9c0f
[ckpt] merge: Merge latest main into yuya/refactor-fused-expert-mappings

yaoyu-33
and
claude
committed
b8c05de
[test] fix: Add missing Llama32ModelProvider1B definition in test_pretrain

yaoyu-33
and
claude
committed
400ef5a
[model] refactor: address PR review comments for MTP disable and expert weight alignment

yaoyu-33
and
claude
committed
55bb430
[ci] fix: Fix ruff lint failures

yaoyu-33
and
claude
committed
a24f701
[test] merge: Resolve conflict in test_pretrain.py with main

yaoyu-33
and
claude
committed
dfa47e4
Merge remote-tracking branch 'origin/main' into yuya/refactor-fused-expert-mappings
yaoyu-33
committed
c4d4484
Merge remote-tracking branch 'origin/yuya/refactor-fused-expert-mappings' into yuya/add-minimax-m2-bridge
yaoyu-33
committed
b85c549

Commits on Mar 18, 2026

Merge origin/main into yuya/add-minimax-m2-bridge

yaoyu-33
and
claude
committed
d04a47e
[misc] merge: Resolve conflicts with origin/yuya/add-minimax-m2-bridge

yaoyu-33
and
claude
committed
e6a2ca9
[build] chore: sync 3rdparty/Megatron-LM submodule to main

yaoyu-33
and
claude
committed
5b68f2f
Merge remote-tracking branch 'origin/main' into yuya/add-minimax-m2-bridge
yaoyu-33
committed
9f7b2b9
[model] revert: restore nemotron_h_provider.py accidentally deleted in merge

yaoyu-33
and
claude
committed
cf38f1e
[model, test, doc] fix: address PR review comments for MiniMax-M2 bridge

yaoyu-33
and
claude
committed
cab0056
[doc] fix: address remaining README review comments for MiniMax-M2

yaoyu-33
and
claude
committed
c31c4a9
[doc] fix: remove CONTAINER_MOUNTS from MiniMax-M2 README

yaoyu-33
and
claude
committed
29a5629
[misc] fix: revert remaining PR review requests from yaoyu-33

yaoyu-33
and
claude
committed
070266a

Commits on Mar 19, 2026

Commits on Mar 20, 2026

[misc] fix: address cuichenx PR review comments for MiniMax-M2

yaoyu-33
and
claude
committed
2c692fe
[ci] fix: resolve merge conflict with main by adopting dynamic test matrix

yaoyu-33
and
claude
committed
02d739b

Commits on Mar 23, 2026

[model] fix: correct MiniMax-M2 HF parameter names in bridge

yaoyu-33
and
claude
committed
e955302
[model] fix: auto-detect MiniMax-M2 HF checkpoint format in bridge

yaoyu-33
and
claude
committed
c599b9c
[model] fix: drop legacy MiniMax-M2 format, use native transformers layout

yaoyu-33
and
claude
committed
3724d87
[model, test] fix: use on-disk block_sparse_moe keys for MiniMax-M2 bridge

yaoyu-33
and
claude
committed
53f2d66
[test] fix: use absolute coverage path in MiniMax-M2 functional test

yaoyu-33
and
claude
committed
c1bfdf2