diff --git a/src/trainer/model.py b/src/trainer/model.py index 49fb424..b48ad4c 100644 --- a/src/trainer/model.py +++ b/src/trainer/model.py @@ -517,7 +517,7 @@ class MoEModel(nn.Module): if monitor: # 使用 eval_loss 作为监控指标 monitor.add_step( - global_step, {"loss": batch_loss_sum, "acc": acc} + global_step, {"loss": batch_loss_sum / (eval_frequency if global_step > 0 else 1), "acc": acc} ) logger.info( f"step: {global_step}, eval_loss: {eval_loss:.4f}, acc: {acc:.4f}, 'batch_loss_sum': {batch_loss_sum / (eval_frequency if global_step > 0 else 1):.4f}, current_lr: {current_lr}"