ros/LlamaFactory
1
0
Fork 0
mirror of https://github.com/hiyouga/LlamaFactory.git synced 2026-03-21 12:03:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor data preprocessing, fix mllm rlhf

Browse Source 
Former-commit-id: 53ff2dd24f9121ea30c95063bb72e49a9b31e980
This commit is contained in:
hiyouga
2024-05-24 04:08:25 +08:00
parent 1078611259
commit bf59383783
15 changed files with 572 additions and 464 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llamafactory/train/orpo/trainer.py
View File

@@ -85,9 +85,7 @@ class CustomORPOTrainer(DPOTrainer):
r"""
Computes the average log probabilities of the labels under the given logits.
"""
all_logits: "torch.Tensor" = model(
input_ids=batch["input_ids"], attention_mask=batch["attention_mask"], return_dict=True, use_cache=False
).logits.to(torch.float32)
all_logits: "torch.Tensor" = model(**batch, return_dict=True, use_cache=False).logits.to(torch.float32)
all_logps = self.get_batch_logps(
logits=all_logits,