From e0bc3c19713c263fb542daef8096dbbb4cf34d7b Mon Sep 17 00:00:00 2001 From: jiaqiw09 <60021713+jiaqiw09@users.noreply.github.com> Date: Thu, 23 Apr 2026 17:29:06 +0800 Subject: [PATCH] [v1] fix epoch and steps (#10422) --- src/llamafactory/v1/core/base_trainer.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/src/llamafactory/v1/core/base_trainer.py b/src/llamafactory/v1/core/base_trainer.py index c3f1d2c69..bac346890 100644 --- a/src/llamafactory/v1/core/base_trainer.py +++ b/src/llamafactory/v1/core/base_trainer.py @@ -227,7 +227,9 @@ class BaseTrainer: """Train the model.""" self.model.train() self.callback_handler.on_train_begin(self.args, self.state) - for epoch in range(self._resume_epoch, self.args.num_train_epochs): + + epoch = self._resume_epoch + while self.global_step < self.num_training_steps: self.state.epoch = epoch self.train_batch_generator.set_epoch(epoch) self.callback_handler.on_epoch_begin(self.args, self.state) @@ -332,6 +334,7 @@ class BaseTrainer: return self.callback_handler.on_epoch_end(self.args, self.state) + epoch += 1 self.callback_handler.on_train_end(self.args, self.state)