Update transformer_layer_infer.py

Author: flyinglandlord

lightllm/models/gemma3/layer_infer/transformer_layer_infer.py

@@ -126,11 +126,9 @@ def _ffn(self, input, infer_state: LlamaInferStateInfo, layer_weight: Gemma3Tran
 
     def context_forward(self, input_embdings, infer_state: InferStateInfo, layer_weight):
         input_embdings = input_embdings.to(torch.bfloat16)
-        # if self.layer_num_ == 0: print('0: layer_input_before_norm', input_embdings)
         input1 = self._att_norm(input_embdings.view(-1, self.embed_dim_).float(), infer_state, layer_weight).to(
             torch.bfloat16
         )
-        # if self.layer_num_ == 0: print('0: layer_input_after_norm', input1)
         cache_kv = self._pre_cache_kv(infer_state, layer_weight)
         q, cache_kv = self._get_qkv(input1, cache_kv, infer_state, layer_weight)
         input1 = None
@@ -141,20 +139,15 @@ def context_forward(self, input_embdings, infer_state: InferStateInfo, layer_wei
         if self.tp_world_size_ > 1:
             all_reduce(o, op=dist.ReduceOp.SUM, group=infer_state.dist_group, async_op=False)
         o = self._ffn_norm(o.float(), infer_state, layer_weight).to(torch.bfloat16)
-        # if self.layer_num_ == 0: print("0:o_after_norm", o)
         input_embdings.add_(o.view(-1, self.embed_dim_))
         o = None
 
-        # if self.layer_num_ == 0: print("0:ffn_hidden_before_norm", input_embdings)
         input1 = self._pre_feedforward_layernorm(input_embdings.float(), infer_state, layer_weight).to(torch.bfloat16)
-        # if self.layer_num_ == 0: print("0:ffn_hidden_after_norm", input1)
         ffn_out = self._ffn(input1, infer_state, layer_weight)
         input1 = None
         if self.tp_world_size_ > 1:
             all_reduce(ffn_out, op=dist.ReduceOp.SUM, group=infer_state.dist_group, async_op=False)
-        # if self.layer_num_ == 0: print("0:ffn_out", ffn_out)
         ffn_out = self._post_feedforward_layernorm(ffn_out.float(), infer_state, layer_weight).to(torch.bfloat16)
-        # if self.layer_num_ == 0: print("0:ffn_out_after_norm", ffn_out)
         input_embdings.add_(ffn_out.view(-1, self.embed_dim_))
         return input_embdings