optimize qwen2-moe layout

README.md

@@ -23,6 +23,7 @@ vLLM是一个快速且易于使用的LLM推理和服务库,使用PageAttention
 | MiniCPM3ForCausalLM   | MiniCPM3                                                    | Yes | Yes |
 | MixtralForCausalLM    | Mixtral-8x7B,Mixtral-8x7B-Instruct                          | Yes | Yes |
 | TeleChat12BForCausalLM (#TelechatForCausalLM) | TeleChat-12B                        | Yes | Yes |
+| Qwen2MoeForCausalLM                 | Qwen2-57B-A14B,Qwen2-57B-A14B-Instruct        | Yes | Yes |
 | LlavaForConditionalGeneration       | LLaMA,LLaMA-2,LLaMA-3                         | Yes | Yes |
 | Qwen2VLForConditionalGeneration     | Qwen2-VL                                      | Yes | Yes |
 | MiniCPMV                            | MiniCPM-V                                     | Yes | Yes |

vllm/model_executor/model_loader/utils.py

@@ -23,7 +23,7 @@ def get_model_architecture(
         model_config: ModelConfig) -> Tuple[Type[nn.Module], str]:
     architectures = getattr(model_config.hf_config, "architectures", [])
     visions = getattr(model_config.hf_config, "visual", []) or getattr(model_config.hf_config, "vision_config", [])
-    support_nn_architectures = ['LlamaForCausalLM', 'QWenLMHeadModel', 'Qwen2ForCausalLM', 'Qwen2VLForConditionalGeneration', 'ChatGLMModel', 'BaichuanForCausalLM', 'BloomForCausalLM', 'MedusaModel']  
+    support_nn_architectures = ['LlamaForCausalLM', 'QWenLMHeadModel', 'Qwen2ForCausalLM', 'Qwen2MoeForCausalLM', 'Qwen2VLForConditionalGeneration', 'ChatGLMModel', 'BaichuanForCausalLM', 'BloomForCausalLM', 'MedusaModel']  
     if any(arch in architectures for arch in support_nn_architectures): 
         if os.getenv('LLAMA_NN') != '0': 
              if (architectures == ['QWenLMHeadModel'] or architectures == ['ChatGLMModel'] ) and visions != []:

vllm/model_executor/models/qwen2_moe.py

@@ -55,6 +55,11 @@ from vllm.utils import print_warning_once
 
 from .utils import is_pp_missing_parameter, make_layers
 
+import os
+import re
+from vllm import _custom_ops as ops
+from vllm.model_executor.utils import pad_weight, gemm_bank_conf
+
 
 class Qwen2MoeMLP(nn.Module):
 
@@ -390,6 +395,16 @@ class Qwen2MoeForCausalLM(nn.Module):
         self.logits_processor = LogitsProcessor(config.vocab_size)
         self.sampler = Sampler()
         
+        self.quant_method = None
+        if quant_config is not None:
+            self.quant_method=quant_config.get_name()
+            self.quant_config=quant_config
+               
+        self.use_llama_nn = os.environ.get('LLAMA_NN') == '1'
+        self.use_gemm_pad = os.environ.get('GEMM_PAD') == '1'
+        self.use_fa_pad = os.environ.get('FA_PAD') == '1'
+        self.use_awq_pad = os.environ.get('AWQ_PAD') == '1'
+
     def forward(
         self,
         input_ids: torch.Tensor,
@@ -529,3 +544,41 @@ class Qwen2MoeForCausalLM(nn.Module):
                     weight_loader = getattr(param, "weight_loader",
                                             default_weight_loader)
                     weight_loader(param, loaded_weight)
+
+        if self.use_llama_nn and self.quant_method is None:
+            lay_key_words = [
+                "gate_up_proj.weight",
+                "down_proj.weight",
+                "mlp.gate.weight",
+                "self_attn.qkv_proj.weight",
+                "self_attn.o_proj.weight",
+                "lm_head.weight",
+            ]
+            combined_words = "|".join(lay_key_words)
+            
+            lay_qkv_words = ["self_attn.qkv_proj.weight"]   
+            qkv_words = "|".join(lay_qkv_words)  
+            
+            lay_qkv_bias_words = ["self_attn.qkv_proj.bias"]   
+            qkv_bias_words = "|".join(lay_qkv_bias_words) 
+            
+            for layername, weight in params_dict.items():
+                # if self.use_fa_pad and (re.findall(qkv_bias_words, layername)):
+                #     weight.data = pad_weight(weight.data, 32)
+                    
+                matches = re.findall(combined_words, layername)
+                if matches:   
+                    if self.use_gemm_pad and gemm_bank_conf(weight.data.shape[0]):
+                        weight.data = pad_weight(weight.data, 32)  
+                    
+                    # if self.use_fa_pad and (re.findall(qkv_words, layername)):
+                    #     if not gemm_bank_conf(weight.data.shape[0]):
+                    #         weight.data = pad_weight(weight.data, 32)
+                        
+                    _weight = torch.zeros_like(weight.data)
+                    ori_shape =_weight.shape
+                    
+                    ops.trans_w16_gemm(_weight, weight.data, _weight.shape[0], _weight.shape[1])
+                    weight.data.copy_(_weight)
+                    
+                    weight.data=weight.data.reshape(ori_shape[1],-1)
\ No newline at end of file

vllm/model_executor/models/qwen2_vl.py

@@ -1158,17 +1158,17 @@ class Qwen2VLForConditionalGeneration(nn.Module, SupportsMultiModal):
             qkv_bias_words = "|".join(lay_qkv_bias_words) 
             
             for layername, weight in params_dict.items():
-                if self.use_fa_pad and (re.findall(qkv_bias_words, layername)):
-                    weight.data = pad_weight(weight.data, 32)
+                # if self.use_fa_pad and (re.findall(qkv_bias_words, layername)):
+                #     weight.data = pad_weight(weight.data, 32)
                     
                 matches = re.findall(combined_words, layername)
                 if matches:   
                     if self.use_gemm_pad and gemm_bank_conf(weight.data.shape[0]):
                         weight.data = pad_weight(weight.data, 32)  
                     
-                    if self.use_fa_pad and (re.findall(qkv_words, layername)):
-                        if not gemm_bank_conf(weight.data.shape[0]):
-                            weight.data = pad_weight(weight.data, 32)
+                    # if self.use_fa_pad and (re.findall(qkv_words, layername)):
+                    #     if not gemm_bank_conf(weight.data.shape[0]):
+                    #         weight.data = pad_weight(weight.data, 32)
                         
                     _weight = torch.zeros_like(weight.data)
                     ori_shape =_weight.shape