423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 20:00:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

add rlhf-v dataset

Browse Source
This commit is contained in:
hiyouga
2024-09-01 22:57:41 +08:00
parent 55027282cd
commit 8e49940746
13 changed files with 118 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/llamafactory/train/dpo/trainer.py
View File

@@ -176,7 +176,6 @@ class CustomDPOTrainer(DPOTrainer):
batch = {k: v.detach().clone() for k, v in batch.items()} # avoid error
all_logits: "torch.Tensor" = model(**batch, return_dict=True, use_cache=False).logits.to(torch.float32)
all_logps, valid_length = get_batch_logps(logits=all_logits, labels=batch["labels"])
if self.loss_type in ["ipo", "orpo", "simpo"]:
all_logps = all_logps / valid_length