genesis-2-rlhf-code / train_dpo.py

Commit History

Skip eval, disable MoE aux loss, disable dropout to fix OOM
9d44c0c
verified

jacobeen06 commited on

Reduce context 4096->3072, add expandable_segments for OOM fix
bff90ea
verified

jacobeen06 commited on

Remove max_prompt_length (not in TRL 1.9 DPOConfig)
382ee20
verified

jacobeen06 commited on

Use DPOConfig for TRL 1.9 compatibility
6c55bf0
verified

jacobeen06 commited on

Fix DPOTrainer: tokenizer -> processing_class (TRL 1.9)
4225e10
verified

jacobeen06 commited on

Fix target modules: remove gate (custom MoE router)
9af44ff
verified

jacobeen06 commited on

DPO training script
3265eba
verified

jacobeen06 commited on