Add support for MiniMax-M2 #42028
vasqu
merged
46 commits into
huggingface:main
from
rogeryoungh:minimax-m2
update: init m2
53639168
update: docs and config
261fe5cc
update: init minimax-m2 test
ac4613ca
update: fix tests
3421fe74
update: use partial_rotary_factor
3a5df7a9
update: some fix
cb17f622
fix: import Unpack from processing_utils
f6775d80
update: apply suggestions from code review
73904ee7
update: remove MiniMaxM2DecoderLayer and MiniMaxM2MLP
6b7e3973
molbap
commented
on 2025-11-12
update: remove use_qk_norm
15657210
update: remove unused use_qk_norm
fa093014
update: update config and attention
11fdb585
update: add to tokenization_auto and remove unused test
93b7598d
Merge branch 'main' into minimax-m2
5de5179d
update: fix decoder layer and experts
f5219ca7
update: fix docs
ef9a1f98
update: make ci happy
a0eea925
refactor: use mapping
2dbfc3b8
update: remove unused comments
640fb9f1
Merge branch 'main' into minimax-m2
8fe4d2bd
Merge branch 'main' into minimax-m2
17c68c83
update: fix rope_params and router
8ba23a6e
update: remove rope_theta
8a76b781
update: test_load_balancing_loss
bcc0aa12
Merge branch 'main' into minimax-m2
5a82172b
update: docs
47c84e2f
vasqu
commented
on 2025-12-08
update: fix default theta
fdb6807e
update to proper default values, proper config rope, simplified modular
1f07a00a
fix docs
25a8b5c4
modular fixup
7ea9a2b3
Merge branch 'main' into minimax-m2
13ed0be9
review comments
9eebcedf
Merge branch 'main' into minimax-m2
668dbef2
update slow tests
90acf7cc
style
e4463e1b
fp32 strict
fde9591a
revert the flag
7b6824f6
Merge branch 'main' into minimax-m2
dda58d9c
Merge branch 'main' into minimax-m2
04c0b5f3
sync with latest changes
87032353
fixup buffer init
53c19ac8
add cache exception to minimax m2 as we have a naming clash
28f26fd2
fix dtype issue in gate
cca9ee72
Merge branch 'main' into minimax-m2
3d6d6718
lift fp8 test restriction and apply new linter rules
b867712c
update docs
252d02d3
ArthurZucker
approved these changes
on 2026-01-09
Reviewers
Assignees
No one assigned
Labels
New model
Milestone
No milestone
Login to write a write a comment.
Login via GitHub