p04798526
/
LLaMA-Factory-Mirror
1
0
Fork 1
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix bug in data loader, support dpo eval

This commit is contained in:
hiyouga 2023-11-03 00:34:26 +08:00
parent 2b5e33c338
commit b355f6cac9
3 changed files with 9 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/llmtuner/dsets/loader.py
View File

@ -82,8 +82,8 @@ def get_dataset(
assistant_role = msg_list[idx + 1][dataset_attr.role]
else:
if (
msg_list[idx][dataset_attr.query] != user_role
or msg_list[idx+1][dataset_attr.query] != assistant_role
msg_list[idx][dataset_attr.role] != user_role
or msg_list[idx+1][dataset_attr.role] != assistant_role
):
raise ValueError("Only accepts conversation in u/a/u/a/u/a order.")
msg_pairs.append((msg_list[idx][dataset_attr.content], msg_list[idx + 1][dataset_attr.content]))

1
src/llmtuner/tuner/dpo/trainer.py
View File

@ -30,6 +30,7 @@ class CustomDPOTrainer(DPOTrainer):
self.is_encoder_decoder = model.config.is_encoder_decoder
self.ref_model = ref_model
self.use_dpo_data_collator = True # hack to avoid warning
self.generate_during_eval = False # disable at evaluation
self.label_pad_token_id = IGNORE_INDEX
self.padding_value = 0
self.beta = beta

6
src/llmtuner/tuner/dpo/workflow.py
View File

@ -58,3 +58,9 @@ def run_dpo(
trainer.save_model()
if trainer.is_world_process_zero() and model_args.plot_loss:
plot_loss(training_args.output_dir, keys=["loss", "eval_loss"])
# Evaluation
if training_args.do_eval:
metrics = trainer.evaluate(metric_key_prefix="eval")
trainer.log_metrics("eval", metrics)
trainer.save_metrics("eval", metrics)