423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2026-09-14 11:15:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[v1] Support multimodal Ulysses CP and memory-efficient chunk loss for SFT (#10762)

Browse Source
This commit is contained in:
xvxuopop
2026-09-09 19:22:25 +08:00
committed by GitHub
parent 673048c6a5
commit 31078aa10a
15 changed files with 961 additions and 187 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llamafactory/v1/trainers/rm_trainer.py
View File

@@ -78,6 +78,8 @@ class RMTrainer(BaseTrainer):
) -> None:
if args.cp_size > 1:
raise NotImplementedError("RM trainer currently only supports cp_size == 1.")
if args.chunk_loss_size is not None:
raise NotImplementedError("Chunk Loss currently only supports SFT training.")
super().__init__(args, model, renderer, train_dataset, callbacks)