Add Code World Model (CWM) #41199
ArthurZucker
merged
34 commits into
huggingface:main
from
jacobkahn:main
[wip][cwm] Code World Model stubs and setup in HF Transformers
ed328b65
[wip] Get other things working
c6adecb6
[wip] Working
4652931a
Tokenizer pad
20485eb6
fix: cwm window attn
a7bfd9ff
temp remove test
db2da319
temp remove test
7d31c57e
Merge branch 'main' into develop
9bab5a71
Fixes
3b3c9103
Temporarily add auto config remapping option until VLLM 0.11 is out
95c3013c
Fix model type and add layer validation
db58f4f4
Add Code World Model (CWM)
db145779
Lint, remove CwmForSequenceClassification
46c55e29
Lint, tests
05a9fb0d
Remove CwmForSequenceClassification
fb8b7218
Lint
921c4ba7
Remove intermediary layer expors/doc errorss, fix tests
ee19e086
Lint
beaa15fe
run python utils/sort_auto_mappings.py --check_only
c7a4be2f
Remove Cwm processor mapping, get check_repo passing
05e86aa8
Remove CwmTextConfig from test
aa614590
Add docstring for CwmConfig
e83610a0
remove global_window and window_pattern params from config
cc535131
Fix docstrings
a662db17
Revert change to auto docstring util
2f8228db
lint
9eb95a94
Fixes minus test improvements
b885a8b8
Alter tests to simply check logits
b80ff8a9
lint
e729586b
Have slow tests use repo, make CwmPretrainedModel passthrough
d13e5c27
Remove decoder layer implementation, use Llama3Decoder + CwmAttetion
9f2a7ab4
Use linear w/o bias for CwmAttention, add token-level integration test
7802c05f
Don't ignore config attention bias
489a72f0
ArthurZucker
approved these changes
on 2025-10-08
Remove attention bias parameter entirely from config
2bf54c23
disabled auto-merge 349 days ago
Pull Request is not mergeable
Pull Request is not mergeable
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub