🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
[fix] residual step order
jingyaogongf659b55761b7
diff --git a/trainer/train_grpo.py b/trainer/train_grpo.pyindex 21f5d9e..8a37dd6 100755--- a/trainer/train_grpo.py+++ b/trainer/train_grpo.py@@ -144,7 +144,7 @@ def grpo_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_modloss = (policy_loss + aux_loss) / args.accumulation_steps # scalarloss.backward()- if step % args.accumulation_steps == 0:+ if step % args.accumulation_steps == 0 or step == iters:if args.grad_clip > 0:torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)optimizer.step()@@ -195,13 +195,6 @@ def grpo_train_epoch(epoch, loader, iters, rollout_engine, ref_model, reward_moddel prompt_inputs, outputs, completion_ids, per_token_logps, ref_per_token_logpsdel completions, rewards, grouped_rewards, mean_r, std_r, advantages, completion_mask, completion_pad_mask, prompt_lens, logp_pos- if step > start_step and step % args.accumulation_steps != 0:- if args.grad_clip > 0:- torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)- optimizer.step()- scheduler.step()- optimizer.zero_grad()-if __name__ == "__main__":parser = argparse.ArgumentParser(description="MiniMind GRPO (Group Relative Policy Optimization)")