add qwen72

lmdeploy/model.py

@@ -448,6 +448,7 @@ If a question does not make any sense, or is not factually coherent, explain why
         return ret
 
 
+@MODELS.register_module(name='qwen-72b')
 @MODELS.register_module(name='qwen-14b')
 @MODELS.register_module(name='qwen-7b')
 class Qwen7BChat(BaseModel):

lmdeploy/serve/turbomind/deploy.py

@@ -864,9 +864,11 @@ def deploy_qwen(model_name: str, model_path: str, tokenizer_path: str,
     def get_tensor(name, trans=True):
         """return a transposed tensor according its name."""
         if trans:
-            return _params[name].cuda().t()
+            # return _params[name].cuda().t()
+            return _params[name].t()
         else:
-            return _params[name].cuda()
+            # return _params[name].cuda()
+            return _params[name]
 
     for i in range(num_layer):
         print(i)