add qwen openapi

llm/LLM.py

@@ -26,7 +26,7 @@ class LLM:
         elif model_name == 'ChatGPT':
             llm = ChatGPT(model_path, api_key=api_key)
         elif model_name == 'Qwen':
-            llm = Qwen(model_path)
+            llm = Qwen(model_path=model_path, api_key=api_key, api_base=proxy_url)
         elif model_name == 'VllmGPT':
             llm = VllmGPT()
         return llm

llm/Qwen.py

@@ -1,12 +1,29 @@
 import os
-import torch
-import requests
-from transformers import AutoModelForCausalLM, AutoTokenizer
+import openai
+
+'''
+`huggingface`连接不上可以使用 `modelscope`
+`pip install modelscope`
+'''
+from modelscope import AutoModelForCausalLM, AutoTokenizer
+#from transformers import AutoModelForCausalLM, AutoTokenizer
+
 os.environ['CUDA_LAUNCH_BLOCKING'] = '1'
 
+
 class Qwen:
-    def __init__(self,  model_path="Qwen/Qwen-1_8B-Chat") -> None:
+    def __init__(self, model_path="Qwen/Qwen-1_8B-Chat", api_base=None, api_key=None) -> None:
         '''暂时不写api版本,与Linly-api相类似,感兴趣可以实现一下'''
+        # 默认本地推理
+        self.local = True
+
+        # api_base和api_key不为空时使用openapi的方式
+        if api_key is not None and base_url is not None:
+            openai.api_base = api_base
+            openai.api_key = api_key
+            self.local = False
+            return
+
         self.model, self.tokenizer = self.init_model(model_path)
         self.data = {}
 
@@ -19,7 +36,20 @@ class Qwen:
         return model, tokenizer
 
     def chat(self, question):
+        # 优先调用qwen openapi的方式
+        if not self.local:
+            # 不使用流式回复的请求
+            response = openai.ChatCompletion.create(
+                model="Qwen",
+                messages=[
+                    {"role": "user", "content": question}
+                ],
+                stream=False,
+                stop=[]
+            )
+            return response.choices[0].message.content
 
+        # 默认本地推理
         self.data["question"] = f"{question} ### Instruction:{question}  ### Response:"
         try:
             response, history = self.model.chat(self.tokenizer, self.data["question"], history=None)
@@ -28,11 +58,12 @@ class Qwen:
         except:
             return "对不起，你的请求出错了，请再次尝试。\nSorry, your request has encountered an error. Please try again.\n"
 
-    
+
 def test():
     llm = Qwen(model_path="Qwen/Qwen-1_8B-Chat")
     answer = llm.chat(question="如何应对压力？")
     print(answer)
 
+
 if __name__ == '__main__':
     test()