Update app.py

app.py

@@ -1,5 +1,52 @@
 import gradio as gr
+from openai import OpenAI
+import os
+import time
 
-demo = gr.load("ambrosfitz/llama-3-history", src="models")
+# Initialize the OpenAI Client with your API key and endpoint
+api_key = os.environ.get("RUNPOD_API_KEY")  # Make sure to set your environment variable correctly
+client = OpenAI(
+    api_key=api_key,
+    base_url="https://api.runpod.ai/v2/vllm-k0g4c60zor9xuu/openai/v1",
+)
 
-demo.launch()
+def get_response(user_message, history):
+    # Format the history for OpenAI
+    history_openai_format = []
+    for human, assistant in history:
+        if human:  # Ensure there's a human message
+            history_openai_format.append({"role": "user", "content": human})
+        if assistant:  # Ensure there's an assistant message
+            history_openai_format.append({"role": "assistant", "content": assistant})
+    history_openai_format.append({"role": "user", "content": user_message})
+
+    # Make the API call
+    response = client.chat.completions.create(
+        model='ambrosfitz/llama-3-history',
+        messages=history_openai_format,
+        temperature=0.5,
+        max_tokens=150
+    )
+
+    # Get the text response
+    bot_message = response.choices[0].message['content'].strip() if response.choices else "No response generated."
+    return bot_message
+
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot()
+    msg = gr.Textbox()
+    clear = gr.Button("Clear")
+
+    def user(user_message, history):
+        if not user_message.strip():  # Handle empty input gracefully
+            return "", history
+        bot_response = get_response(user_message, history)
+        return "", history + [[user_message, bot_response]]
+
+    def clear_chat():
+        return "", []  # Clear the chat history
+
+    msg.submit(user, inputs=[msg, chatbot], outputs=[msg, chatbot])
+    clear.click(clear_chat, inputs=None, outputs=[msg, chatbot])
+
+demo.launch()