Andrew Wu a81feabbd9 DPO transformers v0.29 fixes (#3560) [skip ci] ...

* Deperecate dpo_norm_loss

* Rename chosen/rejected_input_ids to chosen/rejected_ids to match TRL https://github.com/huggingface/trl/pull/5179

* Remove deprecated rpo_alpha

* Remove dead_code tokenize_row

* Add _tokenize override to prevent double bos token on Llama DPO

* Fix DPO loss type now list not string

* Linting fix

* PR fixes

* update _tokenize override for DPO for multimodal

2026-03-31 19:04:53 -04:00

..

callbacks

feat: save checkpoint after training started (#3233)

2025-11-13 10:21:05 -05:00

data

DPO transformers v0.29 fixes (#3560) [skip ci]

2026-03-31 19:04:53 -04:00

lora

feat:merge-lora iterate through bins without loading (#3095)

2026-03-25 08:41:32 -04:00

schemas/validation

automatically enable tf32 if supported (#3473) [skip ci]

2026-03-16 23:47:00 -04:00

test_grpo_rw_fnc.py

feat:openenv rollout_func (#3239) [skip ci]

2025-11-07 08:51:40 -05:00

test_import_helper.py

allow custom trainer_cls to be defined as a module reference in the YAML (#3024) [skip ci]

2025-08-06 22:49:19 -04:00

test_mistral3_processor.py

fix: update MistralProcessor to be v5 compat (#3423)

2026-02-23 11:39:13 -05:00

test_train.py

[GPT-OSS] improve FSDP shard merging and documentation for GPT-OSS (#3073)

2025-08-15 21:25:01 -04:00