修改triton 量化w8a8 config命令方式

vllm/utils/__init__.py

@@ -70,6 +70,7 @@ import vllm.envs as envs
 from vllm.logger import enable_trace_function_call, init_logger
 import json
 
+
 if TYPE_CHECKING:
     from argparse import Namespace
 
@@ -1956,9 +1957,10 @@ 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