423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2026-08-17 13:35:44 +08:00
Code Issues Packages Projects Releases Wiki Activity

[v1] Fix device mesh, fix lora for reward model and fix sp (#10555)

Browse Source
This commit is contained in:
jiaqiw09
2026-06-25 20:05:56 +08:00
committed by GitHub
parent 666ee0ca78
commit b7615dbdc9
7 changed files with 58 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llamafactory/v1/plugins/trainer_plugins/batching.py
View File

@@ -149,8 +149,8 @@ def _pack_padding_free_samples(samples: list[ModelInput], cutoff_len: int) -> Ba
return None
packed["position_ids"] = position_ids
packed["attention_mask"] = [1] * len(position_ids)
return {key: torch.tensor(value).unsqueeze(0) for key, value in packed.items()}
packed["attention_mask"] = None
return {key: None if value is None else torch.tensor(value).unsqueeze(0) for key, value in packed.items()}
@BatchingPlugin("padding_free").register("get_data_provider_batch_size")