Merge 138c367ddb into 21ec325dba

trainer/train_agent.py

@@ -473,8 +473,7 @@ if __name__ == "__main__":
         Logger('torch.compile enabled')
         rollout_engine.update_policy(model)
     if dist.is_initialized():
-        # 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费
-        model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False)
+        model = DistributedDataParallel(model, device_ids=[local_rank])
     rollout_engine.update_policy(model)
 
     for epoch in range(start_epoch, args.epochs):

trainer/train_grpo.py

@@ -314,8 +314,7 @@ if __name__ == "__main__":
         Logger('torch.compile enabled')
         rollout_engine.update_policy(model)
     if dist.is_initialized():
-        # 同 train_ppo：RoPE buffer 各 rank 一致，每步广播纯属浪费
-        model = DistributedDataParallel(model, device_ids=[local_rank], broadcast_buffers=False)
+        model = DistributedDataParallel(model, device_ids=[local_rank])
     rollout_engine.update_policy(model)
     
     # ========== 8. 开始训练 ==========

trainer/train_ppo.py

@@ -36,7 +36,7 @@ def rep_penalty(text, n=3, cap=0.5):
 class CriticModel(MiniMindForCausalLM):
     def __init__(self, params):
         super().__init__(params)
-        # lm_head 不参与 forward，仅靠 tie_word_embeddings 与 embed_tokens 共享权重，解绑后 DDP 会报未使用参数
+        # 替换lm_head为输出单一价值的线性层
         self.value_head = nn.Linear(params.hidden_size, 1)
 
     def forward(self, input_ids=None, attention_mask=None, **kwargs):
@@ -429,9 +429,8 @@ if __name__ == "__main__":
         Logger('torch.compile enabled')
         rollout_engine.update_policy(actor_model)
     if dist.is_initialized():
-        # freqs_cos/freqs_sin 各 rank 由 config 确定性算出，默认每步广播一次纯属浪费
-        actor_model = DistributedDataParallel(actor_model, device_ids=[local_rank], broadcast_buffers=False)
-        critic_model = DistributedDataParallel(critic_model, device_ids=[local_rank], broadcast_buffers=False)
+        actor_model = DistributedDataParallel(actor_model, device_ids=[local_rank])
+        critic_model = DistributedDataParallel(critic_model, device_ids=[local_rank])
     rollout_engine.update_policy(actor_model)
     
     # ========== 8. 开始训练 ==========