Merge branch 'frontier' into production

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