修改triton 量化w8a8 config命令方式

add deps of pd-pp

requirements/rocm.txt

@@ -24,6 +24,9 @@ numa
 pytrie
 setuptools_scm>=8
 cmake==3.29
+quart
+fastrlock==0.8.3
+cupy==12.3.0
 
 torch == 2.5.1
 triton == 3.1

vllm/utils/__init__.py

@@ -2309,9 +2309,10 @@ class W8a8GetCacheJSON:
         self.triton_json_list=[]
         self.weight_shapes=[]
         self.moe_weight_shapes=[]
-        device_name = current_platform.get_device_name().replace(" ", "_")
-        if 'K100_AI' in device_name and torch.cuda.get_device_properties(torch.cuda.current_device()).multi_processor_count == 120:
-            device_name='K100_AI_120'
+        arch_name = torch.cuda.get_device_properties("cuda").gcnArchName.split(':')[0]
+        arch_cu = torch.cuda.get_device_properties(torch.cuda.current_device()).multi_processor_count
+        
+        device_name =arch_name+'_'+str(arch_cu)+'cu'
         self.device_name=device_name
         self.topk=1
         self.quant_method=None