Merge branch 'main' of https://github.com/GaiZhenbiao/ChuanhuChatGPT

modules/models/OpenAI.py

@@ -149,13 +149,13 @@ class OpenAIClient(BaseLLMModel):
             timeout = TIMEOUT_ALL
 
         # 如果有自定义的api-host，使用自定义host发送请求，否则使用默认设置发送请求
-        if shared.state.completion_url != COMPLETION_URL:
-            logging.debug(f"使用自定义API URL: {shared.state.completion_url}")
+        if shared.state.chat_completion_url != CHAT_COMPLETION_URL:
+            logging.debug(f"使用自定义API URL: {shared.state.chat_completion_url}")
 
         with retrieve_proxy():
             try:
                 response = requests.post(
-                    shared.state.completion_url,
+                    shared.state.chat_completion_url,
                     headers=headers,
                     json=payload,
                     stream=stream,
@@ -201,15 +201,23 @@ class OpenAIClient(BaseLLMModel):
                     print(i18n("JSON解析错误,收到的内容: ") + f"{chunk}")
                     error_msg += chunk
                     continue
-                if chunk_length > 6 and "delta" in chunk["choices"][0]:
-                    if chunk["choices"][0]["finish_reason"] == "stop":
-                        break
-                    try:
-                        yield chunk["choices"][0]["delta"]["content"]
-                    except Exception as e:
-                        # logging.error(f"Error: {e}")
-                        continue
-        if error_msg:
+                try:
+                    if chunk_length > 6 and "delta" in chunk["choices"][0]:
+                        if "finish_reason" in chunk["choices"][0]:
+                            finish_reason = chunk["choices"][0]["finish_reason"]
+                        else:
+                            finish_reason = chunk["finish_reason"]
+                        if finish_reason == "stop":
+                            break
+                        try:
+                            yield chunk["choices"][0]["delta"]["content"]
+                        except Exception as e:
+                            # logging.error(f"Error: {e}")
+                            continue
+                except:
+                    print(f"ERROR: {chunk}")
+                    continue
+        if error_msg and not error_msg=="data: [DONE]":
             raise Exception(error_msg)
 
     def set_key(self, new_access_key):
@@ -229,12 +237,12 @@ class OpenAIClient(BaseLLMModel):
             "messages": history,
         }
         # 如果有自定义的api-host，使用自定义host发送请求，否则使用默认设置发送请求
-        if shared.state.completion_url != COMPLETION_URL:
-            logging.debug(f"使用自定义API URL: {shared.state.completion_url}")
+        if shared.state.chat_completion_url != CHAT_COMPLETION_URL:
+            logging.debug(f"使用自定义API URL: {shared.state.chat_completion_url}")
 
         with retrieve_proxy():
             response = requests.post(
-                shared.state.completion_url,
+                shared.state.chat_completion_url,
                 headers=headers,
                 json=payload,
                 stream=False,
@@ -245,7 +253,7 @@ class OpenAIClient(BaseLLMModel):
 
 
     def auto_name_chat_history(self, name_chat_method, user_question, chatbot, user_name, single_turn_checkbox):
-        if len(self.history) == 2 and not single_turn_checkbox:
+        if len(self.history) == 2 and not single_turn_checkbox and not hide_history_when_not_logged_in:
             user_question = self.history[0]["content"]
             if name_chat_method == i18n("模型自动总结（消耗tokens）"):
                 ai_answer = self.history[1]["content"]

modules/models/OpenAIInstruct.py

@@ -0,0 +1,27 @@
+import openai
+from .base_model import BaseLLMModel
+from .. import shared
+from ..config import retrieve_proxy
+
+
+class OpenAI_Instruct_Client(BaseLLMModel):
+    def __init__(self, model_name, api_key, user_name="") -> None:
+        super().__init__(model_name=model_name, user=user_name)
+        self.api_key = api_key
+
+    def _get_instruct_style_input(self):
+        return "\n\n".join([item["content"] for item in self.history])
+
+    @shared.state.switching_api_key
+    def get_answer_at_once(self):
+        prompt = self._get_instruct_style_input()
+        with retrieve_proxy():
+            response = openai.Completion.create(
+                api_key=self.api_key,
+                api_base=shared.state.openai_api_base,
+                model=self.model_name,
+                prompt=prompt,
+                temperature=self.temperature,
+                top_p=self.top_p,
+                )
+        return response.choices[0].text.strip(), response.usage["total_tokens"]

modules/models/base_model.py

@@ -144,13 +144,17 @@ class ModelType(Enum):
     LangchainChat = 10
     Midjourney = 11
     Spark = 12
+    OpenAIInstruct = 13
 
     @classmethod
     def get_type(cls, model_name: str):
         model_type = None
         model_name_lower = model_name.lower()
         if "gpt" in model_name_lower:
-            model_type = ModelType.OpenAI
+            if "instruct" in model_name_lower:
+                model_type = ModelType.OpenAIInstruct
+            else:
+                model_type = ModelType.OpenAI
         elif "chatglm" in model_name_lower:
             model_type = ModelType.ChatGLM
         elif "llama" in model_name_lower or "alpaca" in model_name_lower:
@@ -247,7 +251,7 @@ class BaseLLMModel:
 
     def billing_info(self):
         """get billing infomation, inplement if needed"""
-        logging.warning("billing info not implemented, using default")
+        # logging.warning("billing info not implemented, using default")
         return BILLING_NOT_APPLICABLE_MSG
 
     def count_token(self, user_input):
@@ -743,6 +747,10 @@ class BaseLLMModel:
                     logging.info(new_history)
             except:
                 pass
+            if len(json_s["chatbot"]) < len(json_s["history"]):
+                logging.info("Trimming corrupted history...")
+                json_s["history"] = json_s["history"][-len(json_s["chatbot"]):]
+                logging.info(f"Trimmed history: {json_s['history']}")
             logging.debug(f"{self.user_identifier} 加载对话历史完毕")
             self.history = json_s["history"]
             return os.path.basename(self.history_file_path), json_s["system"], json_s["chatbot"]

modules/models/models.py

@@ -47,6 +47,12 @@ def get_model(
                 top_p=top_p,
                 user_name=user_name,
             )
+        elif model_type == ModelType.OpenAIInstruct:
+            logging.info(f"正在加载OpenAI Instruct模型: {model_name}")
+            from .OpenAIInstruct import OpenAI_Instruct_Client
+            access_key = os.environ.get("OPENAI_API_KEY", access_key)
+            model = OpenAI_Instruct_Client(
+                model_name, api_key=access_key, user_name=user_name)
         elif model_type == ModelType.ChatGLM:
             logging.info(f"正在加载ChatGLM模型: {model_name}")
             from .ChatGLM import ChatGLM_Client

modules/presets.py

@@ -14,7 +14,9 @@ LLAMA_INFERENCER = None
 # ChatGPT 设置
 INITIAL_SYSTEM_PROMPT = "You are a helpful assistant."
 API_HOST = "api.openai.com"
-COMPLETION_URL = "https://api.openai.com/v1/chat/completions"
+OPENAI_API_BASE = "https://api.openai.com/v1"
+CHAT_COMPLETION_URL = "https://api.openai.com/v1/chat/completions"
+COMPLETION_URL = "https://api.openai.com/v1/completions"
 BALANCE_API_URL="https://api.openai.com/dashboard/billing/credit_grants"
 USAGE_API_URL="https://api.openai.com/dashboard/billing/usage"
 HISTORY_DIR = Path("history")
@@ -50,10 +52,11 @@ CHUANHU_DESCRIPTION = i18n("由Bilibili [土川虎虎虎](https://space.bilibili
 
 ONLINE_MODELS = [
     "gpt-3.5-turbo",
+    "gpt-3.5-turbo-instruct",
     "gpt-3.5-turbo-16k",
+    "gpt-4",
     "gpt-3.5-turbo-0301",
     "gpt-3.5-turbo-0613",
-    "gpt-4",
     "gpt-4-0314",
     "gpt-4-0613",
     "gpt-4-32k",

modules/shared.py

@@ -1,4 +1,4 @@
-from modules.presets import COMPLETION_URL, BALANCE_API_URL, USAGE_API_URL, API_HOST
+from modules.presets import CHAT_COMPLETION_URL, BALANCE_API_URL, USAGE_API_URL, API_HOST, OPENAI_API_BASE
 import os
 import queue
 import openai
@@ -6,9 +6,10 @@ import openai
 class State:
     interrupted = False
     multi_api_key = False
-    completion_url = COMPLETION_URL
+    chat_completion_url = CHAT_COMPLETION_URL
     balance_api_url = BALANCE_API_URL
     usage_api_url = USAGE_API_URL
+    openai_api_base = OPENAI_API_BASE
 
     def interrupt(self):
         self.interrupted = True
@@ -22,13 +23,14 @@ class State:
             api_host = f"https://{api_host}"
         if api_host.endswith("/v1"):
             api_host = api_host[:-3]
-        self.completion_url = f"{api_host}/v1/chat/completions"
+        self.chat_completion_url = f"{api_host}/v1/chat/completions"
+        self.openai_api_base = f"{api_host}/v1"
         self.balance_api_url = f"{api_host}/dashboard/billing/credit_grants"
         self.usage_api_url = f"{api_host}/dashboard/billing/usage"
         os.environ["OPENAI_API_BASE"] = api_host
 
     def reset_api_host(self):
-        self.completion_url = COMPLETION_URL
+        self.chat_completion_url = CHAT_COMPLETION_URL
         self.balance_api_url = BALANCE_API_URL
         self.usage_api_url = USAGE_API_URL
         os.environ["OPENAI_API_BASE"] = f"https://{API_HOST}"
@@ -36,7 +38,7 @@ class State:
 
     def reset_all(self):
         self.interrupted = False
-        self.completion_url = COMPLETION_URL
+        self.chat_completion_url = CHAT_COMPLETION_URL
 
     def set_api_key_queue(self, api_key_list):
         self.multi_api_key = True

modules/utils.py

@@ -77,9 +77,6 @@ def auto_name_chat_history(current_model, *args):
 def export_markdown(current_model, *args):
     return current_model.export_markdown(*args)
 
-def load_chat_history(current_model, *args):
-    return current_model.load_chat_history(*args)
-
 def upload_chat_history(current_model, *args):
     return current_model.load_chat_history(*args)
 
@@ -335,6 +332,8 @@ def construct_assistant(text):
 
 def save_file(filename, system, history, chatbot, user_name):
     os.makedirs(os.path.join(HISTORY_DIR, user_name), exist_ok=True)
+    if filename is None:
+        filename = new_auto_history_filename(user_name)
     if filename.endswith(".md"):
         filename = filename[:-3]
     if not filename.endswith(".json") and not filename.endswith(".md"):