Merge pull request #1356 from alphaply/update-for-qwen

修复了qwen使用本地模型时候的报错

config.py

@@ -15,13 +15,13 @@ API_KEY = "此处填API密钥"    # 可同时填写多个API-KEY，用英文逗
 USE_PROXY = False
 if USE_PROXY:
     """
+    代理网络的地址，打开你的代理软件查看代理协议(socks5h / http)、地址(localhost)和端口(11284)
     填写格式是 [协议]://  [地址] :[端口]，填写之前不要忘记把USE_PROXY改成True，如果直接在海外服务器部署，此处不修改
             <配置教程&视频教程> https://github.com/binary-husky/gpt_academic/issues/1>
     [协议] 常见协议无非socks5h/http; 例如 v2**y 和 ss* 的默认本地协议是socks5h; 而cl**h 的默认本地协议是http
-    [地址] 懂的都懂，不懂就填localhost或者127.0.0.1肯定错不了（localhost意思是代理软件安装在本机上）
+    [地址] 填localhost或者127.0.0.1（localhost意思是代理软件安装在本机上）
     [端口] 在代理软件的设置里找。虽然不同的代理软件界面不一样，但端口号都应该在最显眼的位置上
     """
-    # 代理网络的地址，打开你的*学*网软件查看代理的协议(socks5h / http)、地址(localhost)和端口(11284)
     proxies = {
         #          [协议]://  [地址]  :[端口]
         "http":  "socks5h://localhost:11284",  # 再例如  "http":  "http://127.0.0.1:7890",
@@ -100,6 +100,12 @@ AVAIL_LLM_MODELS = ["gpt-3.5-turbo-1106","gpt-4-1106-preview","gpt-4-vision-prev
 MULTI_QUERY_LLM_MODELS = "gpt-3.5-turbo&chatglm3"
 
 
+# 选择本地模型变体（只有当AVAIL_LLM_MODELS包含了对应本地模型时，才会起作用）
+# 如果你选择Qwen系列的模型，那么请在下面的QWEN_MODEL_SELECTION中指定具体的模型
+# 也可以是具体的模型路径
+QWEN_MODEL_SELECTION = "Qwen/Qwen-1_8B-Chat-Int8"
+
+
 # 百度千帆（LLM_MODEL="qianfan"）
 BAIDU_CLOUD_API_KEY = ''
 BAIDU_CLOUD_SECRET_KEY = ''

request_llms/bridge_qwen.py

@@ -1,13 +1,7 @@
 model_name = "Qwen"
 cmd_to_install = "`pip install -r request_llms/requirements_qwen.txt`"
 
-
-from transformers import AutoModel, AutoTokenizer
-import time
-import threading
-import importlib
-from toolbox import update_ui, get_conf, ProxyNetworkActivate
-from multiprocessing import Process, Pipe
+from toolbox import ProxyNetworkActivate, get_conf
 from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
 
 
@@ -24,16 +18,14 @@ class GetQwenLMHandle(LocalLLMHandle):
 
     def load_model_and_tokenizer(self):
         # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行
-        import os, glob
-        import os
-        import platform
-        from modelscope import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
-
+        # from modelscope import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
+        from transformers import AutoModelForCausalLM, AutoTokenizer
+        from transformers.generation import GenerationConfig
         with ProxyNetworkActivate('Download_LLM'):
-            model_id = 'qwen/Qwen-7B-Chat'
-            self._tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen-7B-Chat', trust_remote_code=True, resume_download=True)
+            model_id = get_conf('QWEN_MODEL_SELECTION')
+            self._tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, resume_download=True)
             # use fp16
-            model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True, fp16=True).eval()
+            model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True).eval()
             model.generation_config = GenerationConfig.from_pretrained(model_id, trust_remote_code=True)  # 可指定不同的生成长度、top_p等相关超参
             self._model = model
 
@@ -51,7 +43,7 @@ class GetQwenLMHandle(LocalLLMHandle):
 
         query, max_length, top_p, temperature, history = adaptor(kwargs)
 
-        for response in self._model.chat(self._tokenizer, query, history=history, stream=True):
+        for response in self._model.chat_stream(self._tokenizer, query, history=history):
             yield response
         
     def try_to_import_special_deps(self, **kwargs):

request_llms/requirements_qwen.txt

@@ -1,2 +1,4 @@
 modelscope
-transformers_stream_generator
+transformers_stream_generator
+auto-gptq
+optimum

tests/test_llms.py

@@ -16,8 +16,9 @@ if __name__ == "__main__":
     # from request_llms.bridge_jittorllms_llama import predict_no_ui_long_connection
     # from request_llms.bridge_claude import predict_no_ui_long_connection
     # from request_llms.bridge_internlm import predict_no_ui_long_connection
-    from request_llms.bridge_deepseekcoder import predict_no_ui_long_connection
-    # from request_llms.bridge_qwen import predict_no_ui_long_connection
+    # from request_llms.bridge_deepseekcoder import predict_no_ui_long_connection
+    # from request_llms.bridge_qwen_7B import predict_no_ui_long_connection
+    from request_llms.bridge_qwen import predict_no_ui_long_connection
     # from request_llms.bridge_spark import predict_no_ui_long_connection
     # from request_llms.bridge_zhipu import predict_no_ui_long_connection
     # from request_llms.bridge_chatglm3 import predict_no_ui_long_connection