423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-10-15 08:08:09 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix #850

Browse Source 
Former-commit-id: e5975c4c6b8bd47ec506b0d4a4703bee05495436
This commit is contained in:
hiyouga 2023-09-10 14:22:03 +08:00
parent f91c5f2638
commit b481ad58e6
1 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/llmtuner/tuner/sft/trainer.py
View File

@ -50,9 +50,11 @@ class Seq2SeqPeftTrainer(PeftTrainer):
loss, generated_tokens, labels = super().prediction_step(
model, inputs, prediction_loss_only=prediction_loss_only, ignore_keys=ignore_keys
)
generated_tokens = (
generated_tokens[:, max(prompt_len, label_len):] if generated_tokens is not None else None
)
if generated_tokens is not None:
generated_tokens[:, :max(prompt_len, label_len)] = (
self.tokenizer.pad_token_id * torch.ones_like(generated_tokens[:, :max(prompt_len, label_len)])
)
generated_tokens = generated_tokens.contiguous()
return loss, generated_tokens, labels