From 5c5cb1863b86d92f56ef11d883a0cdeedddc5448 Mon Sep 17 00:00:00 2001 From: YeAnbang Date: Mon, 21 Jul 2025 18:04:20 +0800 Subject: [PATCH] hotfix --- .../coati/distributed/grpo_consumer.py | 14 -------------- 1 file changed, 14 deletions(-) diff --git a/applications/ColossalChat/coati/distributed/grpo_consumer.py b/applications/ColossalChat/coati/distributed/grpo_consumer.py index 7b4843b08..a3f1a1cbb 100644 --- a/applications/ColossalChat/coati/distributed/grpo_consumer.py +++ b/applications/ColossalChat/coati/distributed/grpo_consumer.py @@ -371,20 +371,6 @@ class GRPOConsumer(BaseConsumer): kl = all_reduce_mean(torch.mean(torch.stack(kl)).to(loss.device), self.plugin).data mean_kl.append(kl) mean_loss.append(all_reduce_mean(loss, self.plugin).data) - mini_batch_entropies.append( - all_reduce_mean( - ( - ( - ( - entropy_from_logits(policy_model_logits[:, -num_action:]) - * action_mask_forward_micro_batch - ).sum(-1) - ) - / action_mask_forward_micro_batch.sum(-1) - ).detach(), - self.plugin, - ) - ) else: policy_model_logits = self.policy_model( input_ids=input_ids_forward_micro_batch,