Add Mistral Large 4 #50121
Cyrilvallez
merged
66 commits into
main
from
fix-indexer-comments
github-actions
commented
on 2026-10-29
remi-or
commented
on 2026-10-29
Mappings
2768d604
Modular converter tmp fix
5c2e60ea
Modular first draft
7672d0f4
Fix Rope modular
99f7cf91
Indexer
fd6501c1
Fix pretrained and modular
f95ebe5c
Fix dims
6b2c1131
Conversion mappings
341f681e
Nits
66af45e0
Thread shared_kv_...
2ef552bf
Fix softmaxscaling
da4deb8c
Missing layer type
00afb2d9
Nope in indexer fix
29a525a2
Missing cache
f2b19c5c
Config update
c39e97e7
Conversion, start
80894d78
Inversed mapping
8a8d4064
Fix config
c798d737
Next step in converter
37e8d475
Fix kv sgaring
155390bd
softcapping
c58fd9c9
missing converter
4283b227
Add quantization conversion
562d1240
processor
961cba26
Conversion script
8074db5a
Fix conversions for permuted rope
8cfbe823
Nits
4032be55
Conversion otkenizer ihappens before config
9f048091
Fixes
114ba2ca
Fix routing not in fp32
88f4e162
Resolve some TODOs
dee607a8
Re-wire indexer
fd19ab19
Reorganization
30aae9c0
Modular nits
01ecaad6
Style
a3d18a6d
Fix
a2c3de23
Fix repo
149cfa3f
Doc
c202f0c9
Review 1/n
af22f30b
Disable test of text model only
c8688c74
Fix modular converter broken in rebase and run it
0b25e420
Update to fit new cfg setup
775d7984
Remove norm_topk_prob and add the gate weights to the keep_in_fp32 li…
7a1d0843
Refactor conversion script
f6f2ee55
Fix RoPE conversion bug
da9e1c95
Fix different name scales bug
71b7673b
Review
d71e4616
Fix for a problem that will never happen
b4ba8f3c
Tests, lacking integration tests
395f5062
Fix missing router logits
532601ad
Move logits softcapping o the text config
9598ffb2
Style
21320906
Fix repo
24c8070a
Init alpha to zeros
21f82b88
revert cache -> crop should not be called of uninitialized
9afbaee3
use consolidated.safetensors natively, no new index
89a3fce7
add dtype to config for auto dtype
d28aa359
fix
63927e0c
do not shard the attn for now
1ce4ec18
remove the hack, I fixed it upstream
452c4d84
MAke alpha a buffer
537acd57
fix exporter tests
dfd3e69d
Nits
a274e1d9
ignore
af1cf77b
style
22bbd45e
this was upstreamed
c8085cba
Cyrilvallez
changed the title 
Paullevey
approved these changes
on 2026-10-30
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub