Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix issues of KTOTrainer #1840

Merged
merged 2 commits into from
Jul 17, 2024
Merged

Fix issues of KTOTrainer #1840

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ca5d9bb
Fix issues of KTOTrainer
MAOJIASONG Jul 17, 2024
4faacae
Update trl/trainer/kto_trainer.py
kashif Jul 17, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions trl/trainer/kto_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1350,11 +1350,13 @@ def get_batch_loss_metrics(
if all_num_chosen > 0:
metrics["rewards/chosen_sum"] = self.accelerator.gather(chosen_rewards.nansum()).nansum().item()
metrics["logps/chosen_sum"] = self.accelerator.gather(policy_chosen_logps.nansum()).nansum().item()
metrics["logits/chosen"] = self.accelerator.gather(policy_chosen_logits.nansum()).nanmean().item()
metrics["count/chosen"] = all_num_chosen

if all_num_rejected > 0:
metrics["rewards/rejected_sum"] = self.accelerator.gather(rejected_rewards.nansum()).nansum().item()
metrics["logps/rejected_sum"] = self.accelerator.gather(policy_rejected_logps.nansum()).nansum().item()
metrics["logits/rejected"] = self.accelerator.gather(policy_rejected_logits.nansum()).nanmean().item()
metrics["count/rejected"] = all_num_rejected

metrics["kl"] = kl.item()
Expand Down Expand Up @@ -1512,10 +1514,10 @@ def evaluation_loop(
random_batch = self.data_collator(random_batch_dataset)
random_batch = self._prepare_inputs(random_batch)

target_indicies = [i for i in range(len(random_batch["kl"])) if random_batch["kl"][i] is False]
target_indicies = [i for i in range(len(random_batch["label"])) if random_batch["label"][i] is False]
target_batch = {
"prompt_input_ids": itemgetter(*target_indicies)(random_batch["prompt_input_ids"]),
"prompt_attention_mask": itemgetter(*target_indicies)(random_batch["prompt_attention_mask"]),
"prompt_input_ids": random_batch["prompt_input_ids"][target_indicies],
"prompt_attention_mask": random_batch["prompt_attention_mask"][target_indicies],
"prompt": itemgetter(*target_indicies)(random_batch["prompt"]),
}
policy_output_decoded, ref_output_decoded = self.get_batch_samples(self.model, target_batch)
Expand Down
Loading