Update app.py

app.py

@@ -1,10 +1,9 @@
-import openai
 import time
 import logging
 import gradio as gr
-import os
+
 from src.llm_boilers import llm_boiler
-import configparser
+
 
 logging.basicConfig(format="%(asctime)s - %(message)s", level=logging.INFO)
 logging.warning("READY. App started...")
@@ -12,24 +11,29 @@ 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 = "system\n{}\n"
+    system_format = "<|im_start|>system\n{}<|im_end|>\n"
 
-    def __init__(self, system: str = None, user: str = None, assistant: str = None):
+    def __init__(
+        self, system: str = None, user: str = None, assistant: str = None
+    ) -> None:
         if system is not None:
             self.set_system_prompt(system)
         else:
             self.reset_system_prompt()
-        self.user = user if user else "user\n{}\n"
-        self.assistant = assistant if assistant else "assistant\n{}\n"
+        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.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):
+    def history_as_formatted_str(self, system, history) -> str:
         system = self.system_format.format(system)
         text = system + "".join(
             [
@@ -44,18 +48,27 @@ class Chat:
         )
         text += self.user.format(history[-1][0])
         text += self.response_prefix
-
-        # Truncate text if it exceeds the limit
-        if len(text) > 4096:
-            text = text[-4096:]
+        # 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:]
 
         return text
 
     def clear_history(self, history):
         return []
 
-    def turn(self, user_input: str, history):
-        self.user_turn(user_input, history)
+    def turn(self, user_input: str):
+        self.user_turn(user_input)
         return self.bot_turn()
 
     def user_turn(self, user_input: str, history):
@@ -65,6 +78,8 @@ 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:
@@ -76,18 +91,19 @@ class Chat:
 
 
 def call_inf_server(prompt, openai_key):
-    model_id = "gpt-3.5-turbo"
+    model_id = "eolionross/gpt-3.5-turbo-demo"  # "gpt-3.5-turbo-16k",
     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:
-        # Wait and try one more time
+        # assume it is our error
+        # just wait and try one more time
         print(e)
         time.sleep(2)
         response = model.run(prompt, temperature=1.0)
@@ -95,21 +111,24 @@ def call_inf_server(prompt, openai_key):
         return response
 
 
-
-
-# Get the OpenAI key from the environment variable
-openai_key = os.getenv("OPENAI_API_KEY")
-
-
-
-with gr.Blocks(theme='HaleyCH/HaleyCH_Theme') as demo:
+with gr.Blocks(theme='ParityError/Anime') 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. 
-        """
+        """<h3><center>Chat with gpt-3.5-turbo</center></h3>
+        
+"""
     )
     conversation = Chat()
+    with gr.Row():
+        with gr.Column():
+            # to do: change to openaikey input for public release
+            openai_key = gr.Textbox(
+                label="Open Ai key",
+                value="",
+                type="password",
+                placeholder="Enter your Open ai API KEY here ",
+                info="",
+               
+            )
     chatbot = gr.Chatbot().style(height=400)
     with gr.Row():
         with gr.Column():
@@ -138,48 +157,42 @@ 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.",
             elem_classes=["disclaimer"],
         )
-    with gr.Row():
-        gr.Markdown(
-           "[Privacy policy](https://gist.github.com/samhavens/c29c68cdcd420a9aa0202d0839876dac)",
-            elem_classes=["disclaimer"],
-        )submit_event = msg.submit(
-    fn=conversation.user_turn,
-    inputs=[msg],
-    outputs=[msg, chatbot],
-    queue=False,
-).then(
-    fn=conversation.bot_turn,
-    inputs=[system, chatbot, openai_key],
-    outputs=[chatbot],
-    queue=True,  # Change `queue=True` to `keep_in_queue=True`
-)
-
-submit_click_event = submit.click(
-    fn=conversation.user_turn,
-    inputs=[msg],
-    outputs=[msg, chatbot],
-    queue=False,
-).then(
-    fn=conversation.bot_turn,
-    inputs=[system, chatbot, openai_key],
-    outputs=[chatbot],
-    queue=True,  # Change `queue=True` to `keep_in_queue=True`
-)
-
-stop.click(
-    fn=None,
-    inputs=None,
-    outputs=None,
-    cancels=[submit_event, submit_click_event],
-    queue=False,
-)
 
+    submit_event = msg.submit(
+        fn=conversation.user_turn,
+        inputs=[msg, chatbot],
+        outputs=[msg, chatbot],
+        queue=False,
+    ).then(
+        fn=conversation.bot_turn,
+        inputs=[system, chatbot, openai_key],
+        outputs=[chatbot],
+        queue=True,
+    )
+    submit_click_event = submit.click(
+        fn=conversation.user_turn,
+        inputs=[msg, chatbot],
+        outputs=[msg, chatbot],
+        queue=False,
+    ).then(
+        fn=conversation.bot_turn,
+        inputs=[system, chatbot, openai_key],
+        outputs=[chatbot],
+        queue=True,
+    )
+    stop.click(
+        fn=None,
+        inputs=None,
+        outputs=None,
+        cancels=[submit_event, submit_click_event],
+        queue=False,
+    )
     clear.click(lambda: None, None, chatbot, queue=False).then(
         fn=conversation.clear_history,
         inputs=[chatbot],
@@ -200,4 +213,4 @@ stop.click(
     )
 
 
-demo.queue(max_size=36, concurrency_count=14).launch(debug=True)
+demo.queue(max_size=36, concurrency_count=14).launch(debug=True)