Update app.py

app.py

@@ -6,46 +6,30 @@ import os
 from src.llm_boilers import llm_boiler
 import configparser
 
-# Read the configuration file
-config = configparser.ConfigParser()
-config.read('config.ini')
-
-# Get the OpenAI key from the configuration file
-openai_key = config.get('Credentials', 'openai_key')
-
-# Use openai_key in your code
-
-
-
 logging.basicConfig(format="%(asctime)s - %(message)s", level=logging.INFO)
 logging.warning("READY. App started...")
 
 
 class Chat:
     default_system_prompt = "A conversation between a user and an LLM-based AI assistant. The assistant gives helpful and honest answers."
-    system_format = "<|im_start|>system\n{}<|im_end|>\n"
+    system_format = "system\n{}\n"
 
-    def __init__(
-        self, system: str = None, user: str = None, assistant: str = None
-    ) -> None:
+    def __init__(self, system: str = None, user: str = None, assistant: str = None):
         if system is not None:
             self.set_system_prompt(system)
         else:
             self.reset_system_prompt()
-        self.user = user if user else "<|im_start|>user\n{}<|im_end|>\n"
-        self.assistant = (
-            assistant if assistant else "<|im_start|>assistant\n{}<|im_end|>\n"
-        )
+        self.user = user if user else "user\n{}\n"
+        self.assistant = assistant if assistant else "assistant\n{}\n"
         self.response_prefix = self.assistant.split("{}")[0]
 
     def set_system_prompt(self, system_prompt):
-        # self.system = self.system_format.format(system_prompt)
         return system_prompt
 
     def reset_system_prompt(self):
         return self.set_system_prompt(self.default_system_prompt)
 
-    def history_as_formatted_str(self, system, history) -> str:
+    def history_as_formatted_str(self, system, history):
         system = self.system_format.format(system)
         text = system + "".join(
             [
@@ -60,27 +44,18 @@ class Chat:
         )
         text += self.user.format(history[-1][0])
         text += self.response_prefix
-        # stopgap solution to too long sequences
-        if len(text) > 4500:
-            # delete from the middle between <|im_start|> and <|im_end|>
-            # find the middle ones, then expand out
-            start = text.find("<|im_start|>", 139)
-            end = text.find("<|im_end|>", 139)
-            while end < len(text) and len(text) > 4500:
-                end = text.find("<|im_end|>", end + 1)
-                text = text[:start] + text[end + 1 :]
-        if len(text) > 4500:
-            # the nice way didn't work, just truncate
-            # deleting the beginning
-            text = text[-4500:]
+
+        # Truncate text if it exceeds the limit
+        if len(text) > 4096:
+            text = text[-4096:]
 
         return text
 
     def clear_history(self, history):
         return []
 
-    def turn(self, user_input: str):
-        self.user_turn(user_input)
+    def turn(self, user_input: str, history):
+        self.user_turn(user_input, history)
         return self.bot_turn()
 
     def user_turn(self, user_input: str, history):
@@ -90,8 +65,6 @@ class Chat:
     def bot_turn(self, system, history, openai_key):
         conversation = self.history_as_formatted_str(system, history)
         assistant_response = call_inf_server(conversation, openai_key)
-        # history[-1][-1] = assistant_response
-        # return history
         history[-1][1] = ""
         for chunk in assistant_response:
             try:
@@ -103,19 +76,18 @@ class Chat:
 
 
 def call_inf_server(prompt, openai_key):
-    model_id = "gpt-3.5-turbo"  # "gpt-3.5-turbo-16k",
+    model_id = "gpt-3.5-turbo"
     model = llm_boiler(model_id, openai_key)
     logging.warning(f'Inf via "{model_id}"" for prompt "{prompt}"')
 
     try:
-        # run text generation
+        # Run text generation
         response = model.run(prompt, temperature=1.0)
         logging.warning(f"Result of text generation: {response}")
         return response
 
     except Exception as e:
-        # assume it is our error
-        # just wait and try one more time
+        # Wait and try one more time
         print(e)
         time.sleep(2)
         response = model.run(prompt, temperature=1.0)
@@ -123,34 +95,22 @@ def call_inf_server(prompt, openai_key):
         return response
 
 
-with gr.Blocks(theme='HaleyCH/HaleyCH_Theme') as demo:
-# org : 
-#theme=gr.themes.Glass(
-    #primary_hue="lime",
-    #secondary_hue="emerald",
-    #neutral_hue="zinc",
-    
+# Read the configuration file
+config = configparser.ConfigParser()
+config.read('config.ini')
+
+# Get the OpenAI key from the configuration file
+openai_key = config.get('Credentials', 'openai_key')
 
 
+with gr.Blocks(theme='HaleyCH/HaleyCH_Theme') as demo:
     gr.Markdown(
         """
         <br><h1><center>Chat with gpt-3.5-turbo</center></h1>
-        This is a lightweight  gpt-3.5-turbo conversation
-        completion. 
-"""
-     )
+        This is a lightweight gpt-3.5-turbo conversation completion. 
+        """
+    )
     conversation = Chat()
-    with gr.Row():
-        with gr.Column():
-            # to do: change to openaikey input for public release
-            openai_key = gr.Textbox(
-                label="OpenAI Key",
-                value="", 
-                type="password",
-                placeholder="os.environ.get('openai_key')",
-                info="You have to provide your own OpenAI API key.",
-            )
-            
     chatbot = gr.Chatbot().style(height=400)
     with gr.Row():
         with gr.Column():
@@ -179,7 +139,7 @@ with gr.Blocks(theme='HaleyCH/HaleyCH_Theme') as demo:
                         reset = gr.Button("Reset System Prompt")
     with gr.Row():
         gr.Markdown(
-            "Disclaimer: The gpt-3.5-turbo model can produce factually incorrect output, and should not be solely relied on to produce "
+            "Disclaimer: The gpt-3.5-turbo model can produce factually incorrect output and should not be solely relied on to produce "
             "factually accurate information. The gpt-3.5-turbo model was trained on various public datasets; while great efforts "
             "have been taken to clean the pretraining data, it is possible that this model could generate lewd, "
             "biased, or otherwise offensive outputs.",
@@ -240,4 +200,4 @@ with gr.Blocks(theme='HaleyCH/HaleyCH_Theme') as demo:
     )
 
 
-demo.queue(max_size=36, concurrency_count=14).launch(debug=True)
+demo.queue(max_size=36, concurrency_count=14).launch(debug=True)