增加AWQ相关环境变量控制

vllm/model_executor/layers/quantization/awq.py

 from typing import Any, Dict, List, Optional
 
 import torch
+import os
 import torch.nn.functional as F
 
 from vllm import _custom_ops as ops
@@ -98,6 +99,8 @@ class AWQLinearMethod(LinearMethodBase):
     def __init__(self, quant_config: AWQConfig):
         self.quant_config = quant_config
         self.awqsingleton= AWQShareWorkSpace()
+        self.use_awq_pad = os.environ.get('AWQ_PAD') == '1'
+        self.AWQ_CK_GEMMBS =int(os.getenv('AWQ_CK_GEMMBS', '20000'))
 
     def create_weights(self, layer: torch.nn.Module,
                        input_size_per_partition: int,
@@ -190,12 +193,15 @@ class AWQLinearMethod(LinearMethodBase):
         k = reshaped_x.shape[-1]
         n = qweight.shape[0]
         
-        if k % 4096==0:
-            padding_group=2
+        if self.use_awq_pad:
+            if k % 4096 == 0:
+                padding_group=2
+            else:
+                padding_group=0
         else:
             padding_group=0
         
-        if m<20000: 
+        if m <= self.AWQ_CK_GEMMBS: 
             out = ops.awq_gemm(reshaped_x,
                             qweight,
                             zeros_and_scales,
@@ -208,7 +214,7 @@ class AWQLinearMethod(LinearMethodBase):
                             self.awqsingleton.awqworkshapcesize)
         else: 
             #下面是采用rocblas的做法
-            deqweight=ops.dequant_w4_gemm_colmajor(    #shape[n,k/8]--->[n,k]
+            deqweight=ops.dequant_w4_gemm_colmajor(    # shape[n, k/8] ---> [n,k]
                             qweight, 
                             zeros_and_scales,
                             k,

vllm/model_executor/model_loader/utils.py

@@ -30,10 +30,22 @@ def get_model_architecture(
             os.environ['GEMM_PAD'] = '0'
         if os.getenv('FA_PAD') != '1': 
             os.environ['FA_PAD'] = '0'
+        try:
+            if os.getenv('AWQ_PAD') == '0' or ((torch.cuda.isCurrentDeviceEco(torch.cuda.current_device())) and os.getenv('AWQ_PAD') == None):
+                os.environ['AWQ_PAD'] = '0'
+            else:
+                os.environ['AWQ_PAD'] = '1'
+        except Exception as e:
+            print("Info: this version torch cannot get eco device info.\n")
+            if os.getenv('AWQ_PAD') != '0': 
+                os.environ['AWQ_PAD'] = '1'
+            else:
+                os.environ['AWQ_PAD'] = '0'
     else:
         os.environ['LLAMA_NN'] = '0'
         os.environ['GEMM_PAD'] = '0'
         os.environ['FA_PAD'] = '0'
+        os.environ['AWQ_PAD'] = '0'
         
     # Special handling for quantized Mixtral.
     # FIXME(woosuk): This is a temporary hack.

vllm/model_executor/models/llama.py

@@ -457,6 +457,7 @@ class LlamaForCausalLM(nn.Module, SupportsLoRA):
         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,
@@ -633,7 +634,7 @@ class LlamaForCausalLM(nn.Module, SupportsLoRA):
                     zeros_and_scalse.data=zeros_and_scalse.reshape(dim_n,-1)    #[k/greop_size,n]------>[n,k/group_size]
                     qweight.data=qweight.data.reshape(dim_n,-1)                      #[k,n/8]---->[n,k/8]  
                 
-                    if dim_k % 4096==0:
+                    if dim_k % 4096==0 and self.use_awq_pad:
                         zeros_and_scalse_pad= torch.zeros(dim_n,pad_group,dtype=torch.int32).cuda()
                         zeros_and_scalse.data=torch.cat((zeros_and_scalse.data,zeros_and_scalse_pad),dim=1).contiguous()
                         qweight_pad= torch.zeros(dim_n,int(group_size//4),dtype=torch.int32).cuda()

vllm/model_executor/models/qwen.py

@@ -903,6 +903,7 @@ class QWenLMHeadModel(nn.Module, SupportsMultiModal):
         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 _get_image_input_type(
             self,
@@ -1085,7 +1086,7 @@ class QWenLMHeadModel(nn.Module, SupportsMultiModal):
                     zeros_and_scalse.data=zeros_and_scalse.reshape(dim_n,-1)    #[k/greop_size,n]------>[n,k/group_size]
                     qweight.data=qweight.data.reshape(dim_n,-1)                      #[k,n/8]---->[n,k/8]  
                 
-                    if dim_k % 4096==0:
+                    if dim_k % 4096==0 and self.use_awq_pad:
                         zeros_and_scalse_pad= torch.zeros(dim_n,pad_group,dtype=torch.int32).cuda()
                         zeros_and_scalse.data=torch.cat((zeros_and_scalse.data,zeros_and_scalse_pad),dim=1).contiguous()
                         qweight_pad= torch.zeros(dim_n,int(group_size//4),dtype=torch.int32).cuda()

vllm/model_executor/models/qwen2.py

@@ -378,6 +378,7 @@ class Qwen2ForCausalLM(nn.Module, SupportsLoRA):
         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,
@@ -537,7 +538,7 @@ class Qwen2ForCausalLM(nn.Module, SupportsLoRA):
                     zeros_and_scalse.data=zeros_and_scalse.reshape(dim_n,-1)    #[k/greop_size,n]------>[n,k/group_size]
                     qweight.data=qweight.data.reshape(dim_n,-1)                      #[k,n/8]---->[n,k/8]  
                 
-                    if dim_k % 4096==0:
+                    if dim_k % 4096==0 and self.use_awq_pad:
                         zeros_and_scalse_pad= torch.zeros(dim_n,pad_group,dtype=torch.int32).cuda()
                         zeros_and_scalse.data=torch.cat((zeros_and_scalse.data,zeros_and_scalse_pad),dim=1).contiguous()
                         qweight_pad= torch.zeros(dim_n,int(group_size//4),dtype=torch.int32).cuda()