增加了一大堆参数控制

ChuanhuChatbot.py

@@ -159,21 +159,74 @@ with gr.Blocks(css=customCSS, theme=small_and_beautiful_theme) as demo:
                     default_btn = gr.Button("🔙 恢复默认设置")
 
                     with gr.Accordion("参数", open=False):
+                        temperature_slider = gr.Slider(
+                            minimum=-0,
+                            maximum=2.0,
+                            value=1.0,
+                            step=0.1,
+                            interactive=True,
+                            label="temperature",
+                        )
                         top_p_slider = gr.Slider(
                             minimum=-0,
                             maximum=1.0,
                             value=1.0,
                             step=0.05,
                             interactive=True,
-                            label="Top-p",
+                            label="top-p",
                         )
-                        temperature_slider = gr.Slider(
-                            minimum=-0,
+                        n_choices_slider = gr.Slider(
+                            minimum=1,
+                            maximum=1,
+                            value=1,
+                            step=1,
+                            interactive=True,
+                            label="n choices",
+                        )
+                        stop_sequence_txt = gr.Textbox(
+                            show_label=True,
+                            placeholder=f"在这里输入停止符，用英文逗号隔开...",
+                            label="stop",
+                            value="",
+                            lines=1,
+                        )
+                        max_tokens_slider = gr.Slider(
+                            minimum=1,
+                            maximum=4096,
+                            value=4096,
+                            step=1,
+                            interactive=True,
+                            label="max tokens",
+                        )
+                        presence_penalty_slider = gr.Slider(
+                            minimum=-2.0,
                             maximum=2.0,
-                            value=1.0,
-                            step=0.1,
+                            value=0.0,
+                            step=0.01,
+                            interactive=True,
+                            label="presence penalty",
+                        )
+                        frequency_penalty_slider = gr.Slider(
+                            minimum=-2.0,
+                            maximum=2.0,
+                            value=0.0,
+                            step=0.01,
                             interactive=True,
-                            label="Temperature",
+                            label="frequency penalty",
+                        )
+                        logit_bias_txt = gr.Textbox(
+                            show_label=True,
+                            placeholder=f"word:likelihood",
+                            label="logit bias",
+                            value="",
+                            lines=1,
+                        )
+                        user = gr.Textbox(
+                            show_label=True,
+                            placeholder=f"用于定位滥用行为",
+                            label="用户名",
+                            value=user_name.value,
+                            lines=1,
                         )
 
                     with gr.Accordion("网络设置", open=False):

modules/base_model.py

@@ -41,19 +41,42 @@ class ModelType(Enum):
 
 
 class BaseLLMModel:
-    def __init__(self, model_name, temperature=1.0, top_p=1.0, max_generation_token=None, system_prompt="") -> None:
+    def __init__(
+        self,
+        model_name,
+        system_prompt="",
+        temperature=1.0,
+        top_p=1.0,
+        n_choices=1,
+        stop=None,
+        max_generation_token=None,
+        presence_penalty=0,
+        frequency_penalty=0,
+        logit_bias=None,
+        user="",
+    ) -> None:
         self.history = []
         self.all_token_counts = []
         self.model_name = model_name
         self.model_type = ModelType.get_type(model_name)
         self.token_upper_limit = MODEL_TOKEN_LIMIT[model_name]
-        self.max_generation_token = max_generation_token if max_generation_token is not None else self.token_upper_limit
         self.interrupted = False
-        self.temperature = temperature
-        self.top_p = top_p
         self.system_prompt = system_prompt
         self.api_key = None
 
+        self.temperature = temperature
+        self.top_p = top_p
+        self.n_choices = n_choices
+        self.stop = stop
+        self.max_generation_token = (
+            max_generation_token
+            if max_generation_token is not None
+            else self.token_upper_limit
+        )
+        self.presence_penalty = presence_penalty
+        self.frequency_penalty = frequency_penalty
+        self.logit_bias = logit_bias
+        self.user = user
 
     def get_answer_stream_iter(self):
         """stream predict, need to be implemented
@@ -75,15 +98,11 @@ class BaseLLMModel:
         """get billing infomation, inplement if needed"""
         return BILLING_NOT_APPLICABLE_MSG
 
-
     def count_token(self, user_input):
-        """get token count from input, implement if needed
-        """
+        """get token count from input, implement if needed"""
         return 0
 
-    def stream_next_chatbot(
-        self, inputs, chatbot, fake_input=None, display_append=""
-    ):
+    def stream_next_chatbot(self, inputs, chatbot, fake_input=None, display_append=""):
         def get_return_value():
             return chatbot, status_text
 
@@ -106,9 +125,7 @@ class BaseLLMModel:
             status_text = self.token_message()
             yield get_return_value()
 
-    def next_chatbot_at_once(
-        self, inputs, chatbot, fake_input=None, display_append=""
-    ):
+    def next_chatbot_at_once(self, inputs, chatbot, fake_input=None, display_append=""):
         if fake_input:
             chatbot.append((fake_input, ""))
         else:
@@ -122,7 +139,7 @@ class BaseLLMModel:
         if fake_input is not None:
             self.history[-2] = construct_user(fake_input)
         self.history[-1] = construct_assistant(ai_reply)
-        chatbot[-1] = (chatbot[-1][0], ai_reply+display_append)
+        chatbot[-1] = (chatbot[-1][0], ai_reply + display_append)
         if fake_input is not None:
             self.all_token_counts[-1] += count_token(construct_assistant(ai_reply))
         else:
@@ -277,12 +294,15 @@ class BaseLLMModel:
             self.history = self.history[-4:]
             self.all_token_counts = self.all_token_counts[-2:]
 
-
         max_token = self.token_upper_limit - TOKEN_OFFSET
 
         if sum(self.all_token_counts) > max_token and should_check_token_count:
             count = 0
-            while sum(self.all_token_counts) > self.token_upper_limit * REDUCE_TOKEN_FACTOR and sum(self.all_token_counts) > 0:
+            while (
+                sum(self.all_token_counts)
+                > self.token_upper_limit * REDUCE_TOKEN_FACTOR
+                and sum(self.all_token_counts) > 0
+            ):
                 count += 1
                 del self.all_token_counts[0]
                 del self.history[:2]
@@ -385,7 +405,7 @@ class BaseLLMModel:
         msg = "删除了一组对话"
         return chatbot, msg
 
-    def token_message(self, token_lst = None):
+    def token_message(self, token_lst=None):
         if token_lst is None:
             token_lst = self.all_token_counts
         token_sum = 0
@@ -433,4 +453,4 @@ class BaseLLMModel:
             return filename, json_s["system"], json_s["chatbot"]
         except FileNotFoundError:
             logging.warning(f"{user_name} 没有找到对话历史文件，不执行任何操作")
-            return filename, self.system_prompt, chatbot
+            return filename, self.system_prompt, chatbot

modules/models.py

@@ -26,16 +26,25 @@ from .base_model import BaseLLMModel, ModelType
 
 class OpenAIClient(BaseLLMModel):
     def __init__(
-        self, model_name, api_key, system_prompt=INITIAL_SYSTEM_PROMPT, temperature=1.0, top_p=1.0
+        self,
+        model_name,
+        api_key,
+        system_prompt=INITIAL_SYSTEM_PROMPT,
+        temperature=1.0,
+        top_p=1.0,
     ) -> None:
-        super().__init__(model_name=model_name, temperature=temperature, top_p=top_p, system_prompt=system_prompt)
+        super().__init__(
+            model_name=model_name,
+            temperature=temperature,
+            top_p=top_p,
+            system_prompt=system_prompt,
+        )
         self.api_key = api_key
         self.headers = {
             "Content-Type": "application/json",
             "Authorization": f"Bearer {self.api_key}",
         }
 
-
     def get_answer_stream_iter(self):
         response = self._get_response(stream=True)
         if response is not None:
@@ -57,7 +66,9 @@ class OpenAIClient(BaseLLMModel):
     def count_token(self, user_input):
         input_token_count = count_token(construct_user(user_input))
         if self.system_prompt is not None and len(self.all_token_counts) == 0:
-            system_prompt_token_count = count_token(construct_system(self.system_prompt))
+            system_prompt_token_count = count_token(
+                construct_system(self.system_prompt)
+            )
             return input_token_count + system_prompt_token_count
         return input_token_count
 
@@ -70,18 +81,20 @@ class OpenAIClient(BaseLLMModel):
             try:
                 usage_data = self._get_billing_data(usage_url)
             except Exception as e:
-                logging.error(f"获取API使用情况失败:"+str(e))
+                logging.error(f"获取API使用情况失败:" + str(e))
                 return f"**获取API使用情况失败**"
-            rounded_usage = "{:.5f}".format(usage_data['total_usage']/100)
+            rounded_usage = "{:.5f}".format(usage_data["total_usage"] / 100)
             return f"**本月使用金额** \u3000 ${rounded_usage}"
         except requests.exceptions.ConnectTimeout:
-            status_text = STANDARD_ERROR_MSG + CONNECTION_TIMEOUT_MSG + ERROR_RETRIEVE_MSG
+            status_text = (
+                STANDARD_ERROR_MSG + CONNECTION_TIMEOUT_MSG + ERROR_RETRIEVE_MSG
+            )
             return status_text
         except requests.exceptions.ReadTimeout:
             status_text = STANDARD_ERROR_MSG + READ_TIMEOUT_MSG + ERROR_RETRIEVE_MSG
             return status_text
         except Exception as e:
-            logging.error(f"获取API使用情况失败:"+str(e))
+            logging.error(f"获取API使用情况失败:" + str(e))
             return STANDARD_ERROR_MSG + ERROR_RETRIEVE_MSG
 
     @shared.state.switching_api_key  # 在不开启多账号模式的时候，这个装饰器不会起作用
@@ -110,6 +123,7 @@ class OpenAIClient(BaseLLMModel):
             "stream": stream,
             "presence_penalty": 0,
             "frequency_penalty": 0,
+            "max_tokens": self.max_generation_token,
         }
         if stream:
             timeout = TIMEOUT_STREAMING
@@ -145,7 +159,9 @@ class OpenAIClient(BaseLLMModel):
             data = response.json()
             return data
         else:
-            raise Exception(f"API request failed with status code {response.status_code}: {response.text}")
+            raise Exception(
+                f"API request failed with status code {response.status_code}: {response.text}"
+            )
 
     def _decode_chat_response(self, response):
         for chunk in response.iter_lines():
@@ -166,15 +182,25 @@ class OpenAIClient(BaseLLMModel):
                         # logging.error(f"Error: {e}")
                         continue
 
-def get_model(model_name, access_key=None, temperature=None, top_p=None, system_prompt = None) -> BaseLLMModel:
+
+def get_model(
+    model_name, access_key=None, temperature=None, top_p=None, system_prompt=None
+) -> BaseLLMModel:
     msg = f"模型设置为了： {model_name}"
     logging.info(msg)
     model_type = ModelType.get_type(model_name)
     if model_type == ModelType.OpenAI:
-        model = OpenAIClient(model_name=model_name, api_key=access_key,system_prompt=system_prompt, temperature=temperature, top_p=top_p)
+        model = OpenAIClient(
+            model_name=model_name,
+            api_key=access_key,
+            system_prompt=system_prompt,
+            temperature=temperature,
+            top_p=top_p,
+        )
     return model, msg
 
-if __name__=="__main__":
+
+if __name__ == "__main__":
     with open("config.json", "r") as f:
         openai_api_key = cjson.load(f)["openai_api_key"]
     client = OpenAIClient("gpt-3.5-turbo", openai_api_key)