Added history and Better UI (#7)

- Update requirements.txt (8186757df9cd4e8c124f9e8d8544bef8c2e323e1)
- Update app.py (11dcd687af7926659ec40b082561a7c9462889f6)


Co-authored-by: Nishith Jain <[email protected]>

app.py

@@ -11,127 +11,113 @@ MODELS = [
     "Meta-Llama-3.1-8B-Instruct"
 ]
 
+# Sambanova API base URL
+API_BASE = "https://api.sambanova.ai/v1"
+
 def create_client(api_key=None):
-    openai.api_base = "https://api.sambanova.ai/v1"  # Fixed Base URL
+    """Creates an OpenAI client instance."""
     if api_key:
         openai.api_key = api_key
     else:
         openai.api_key = os.getenv("API_KEY")
 
-def chat_with_ai(message, chat_history, system_prompt):
-    messages = [
-        {"role": "system", "content": system_prompt},
-    ]
-
-    for human, ai in chat_history:
-        messages.append({"role": "user", "content": human})
-        messages.append({"role": "assistant", "content": ai})
+    return openai.OpenAI(api_key=openai.api_key, base_url=API_BASE)
 
+def chat_with_ai(message, chat_history, system_prompt):
+    """Formats the chat history for the API call."""
+    messages = [{"role": "system", "content": system_prompt}]
+    print(type(chat_history))
+    for tup in chat_history:
+        print(type(tup))
+        first_key = list(tup.keys())[0]  # First key
+        last_key = list(tup.keys())[-1]   # Last key
+        messages.append({"role": "user", "content": tup[first_key]})
+        messages.append({"role": "assistant", "content": tup[last_key]})
     messages.append({"role": "user", "content": message})
-
     return messages
 
 def respond(message, chat_history, model, system_prompt, thinking_budget, api_key):
-    print("Starting respond function...")
-    create_client(api_key)  # Sets api_key and api_base globally
+    """Sends the message to the API and gets the response."""
+    client = create_client(api_key)
     messages = chat_with_ai(message, chat_history, system_prompt.format(budget=thinking_budget))
     start_time = time.time()
 
     try:
-        print("Calling OpenAI API...")
-        completion = openai.ChatCompletion.create(
-            model=model,
-            messages=messages,
-            stream=False  # Set to False for synchronous response
-        )
-        response = completion.choices[0].message['content']
+        completion = client.chat.completions.create(model=model, messages=messages)
+        response = completion.choices[0].message.content
         thinking_time = time.time() - start_time
-        print("Response received from OpenAI API.")
         return response, thinking_time
     except Exception as e:
         error_message = f"Error: {str(e)}"
-        print(error_message)
         return error_message, time.time() - start_time
 
 def parse_response(response):
+    """Parses the response from the API."""
     answer_match = re.search(r'<answer>(.*?)</answer>', response, re.DOTALL)
     reflection_match = re.search(r'<reflection>(.*?)</reflection>', response, re.DOTALL)
 
     answer = answer_match.group(1).strip() if answer_match else ""
     reflection = reflection_match.group(1).strip() if reflection_match else ""
-
     steps = re.findall(r'<step>(.*?)</step>', response, re.DOTALL)
-    if answer is not "":
-        return answer, reflection, steps
-    else:
-        return response, "", ""
 
-def process_chat(message, history, model, system_prompt, thinking_budget, api_key):
-    print(f"Received message: {message}")
-    #if not api_key:
-    #    print("API key missing")
-    #    return history + [("System", "Please provide your API Key before starting the chat.")]
+    return answer, reflection, steps
 
-    try:
-        formatted_system_prompt = system_prompt.format(budget=thinking_budget)
-    except KeyError as e:
-        error_msg = f"System prompt missing placeholder: {str(e)}"
-        print(error_msg)
-        return history + [("System", error_msg)]
-
-    response, thinking_time = respond(message, history, model, formatted_system_prompt, thinking_budget, api_key)
+def generate(message, history, model, system_prompt, thinking_budget, api_key):
+    """Generates the chatbot response."""
+    response, thinking_time = respond(message, history, model, system_prompt, thinking_budget, api_key)
 
     if response.startswith("Error:"):
-        return history + [("System", response)]
+        return history + [({"role": "system", "content": response},)], ""
 
     answer, reflection, steps = parse_response(response)
 
-    formatted_response = f"**Answer:** {answer}\n\n**Reflection:** {reflection}\n\n**Thinking Steps:**\n"
-    for i, step in enumerate(steps, 1):
-        formatted_response += f"**Step {i}:** {step}\n"
+    messages = []
+    messages.append({"role": "user", "content": message})
+
+    formatted_steps = [f"Step {i}: {step}" for i, step in enumerate(steps, 1)]
+    all_steps = "\n".join(formatted_steps) + f"\n\nReflection: {reflection}"
 
-    formatted_response += f"\n**Thinking time:** {thinking_time:.2f} s"
+    messages.append({"role": "assistant", "content": all_steps, "metadata": {"title": f"Thinking Time: {thinking_time:.2f} sec"}})
+    messages.append({"role": "assistant", "content": answer})
 
-    print(f"Appended response: {formatted_response}")
-    return history + [(message, formatted_response)]
+    return history + messages, ""
 
 # Define the default system prompt
-default_system_prompt = """
+DEFAULT_SYSTEM_PROMPT = """
 You are a helpful assistant in normal conversation.
-When given a problem to solve, you are an expert problem-solving assistant. Your task is to provide a detailed, step-by-step solution to a given question. Follow these instructions carefully:
-
+When given a problem to solve, you are an expert problem-solving assistant. 
+Your task is to provide a detailed, step-by-step solution to a given question. 
+Follow these instructions carefully:
 1. Read the given question carefully and reset counter between <count> and </count> to {budget}
 2. Generate a detailed, logical step-by-step solution.
 3. Enclose each step of your solution within <step> and </step> tags.
-4. You are allowed to use at most {budget} steps (starting budget), keep track of it by counting down within tags <count> </count>, STOP GENERATING MORE STEPS when hitting 0, you don't have to use all of them.
-5. Do a self-reflection when you are unsure about how to proceed, based on the self-reflection and reward, decides whether you need to return to the previous steps.
-6. After completing the solution steps, reorganize and synthesize the steps into the final answer within <answer> and </answer> tags.
-7. Provide a critical, honest and subjective self-evaluation of your reasoning process within <reflection> and </reflection> tags.
-8. Assign a quality score to your solution as a float between 0.0 (lowest quality) and 1.0 (highest quality), enclosed in <reward> and </reward> tags.
-
+4. You are allowed to use at most {budget} steps (starting budget), 
+   keep track of it by counting down within tags <count> </count>, 
+   STOP GENERATING MORE STEPS when hitting 0, you don't have to use all of them.
+5. Do a self-reflection when you are unsure about how to proceed, 
+   based on the self-reflection and reward, decides whether you need to return 
+   to the previous steps.
+6. After completing the solution steps, reorganize and synthesize the steps 
+   into the final answer within <answer> and </answer> tags.
+7. Provide a critical, honest and subjective self-evaluation of your reasoning 
+   process within <reflection> and </reflection> tags.
+8. Assign a quality score to your solution as a float between 0.0 (lowest 
+   quality) and 1.0 (highest quality), enclosed in <reward> and </reward> tags.
 Example format:            
 <count> [starting budget] </count>
-
 <step> [Content of step 1] </step>
 <count> [remaining budget] </count>
-
 <step> [Content of step 2] </step>
 <reflection> [Evaluation of the steps so far] </reflection>
 <reward> [Float between 0.0 and 1.0] </reward>
 <count> [remaining budget] </count>
-
 <step> [Content of step 3 or Content of some previous step] </step>
 <count> [remaining budget] </count>
-
 ...
-
 <step>  [Content of final step] </step>
 <count> [remaining budget] </count>
-
-<answer> [Final Answer] </answer>
-
+<answer> [Final Answer] </answer> (must give final answer in this format)
 <reflection> [Evaluation of the solution] </reflection>
-
 <reward> [Float between 0.0 and 1.0] </reward>
 """
 
@@ -140,65 +126,20 @@ with gr.Blocks() as demo:
     gr.Markdown("[Powered by Llama3.1 models through SambaNova Cloud API](https://sambanova.ai/fast-api?api_ref=907266)")
 
     with gr.Row():
-        api_key = gr.Textbox(
-            label="API Key",
-            type="password",
-            placeholder="(Optional) Enter your API key here for more availability"
-        )
+        api_key = gr.Textbox(label="API Key", type="password", placeholder="(Optional) Enter your API key here for more availability")
 
     with gr.Row():
-        model = gr.Dropdown(
-            choices=MODELS,
-            label="Select Model",
-            value=MODELS[0]
-        )
-        thinking_budget = gr.Slider(
-            minimum=1,
-            maximum=100,
-            value=10,
-            step=1,
-            label="Thinking Budget"
-        )
-
-    system_prompt = gr.Textbox(
-        label="System Prompt",
-        value=default_system_prompt,
-        lines=15,
-        interactive=True
-    )
+        model = gr.Dropdown(choices=MODELS, label="Select Model", value=MODELS[0])
+        thinking_budget = gr.Slider(minimum=1, maximum=100, value=10, step=1, label="Thinking Budget", info="maximum times a model can think")
 
-    with gr.Row():
-        msg = gr.Textbox(
-            label="Type your message here...",
-            placeholder="Enter your message..."
-        )
-        submit = gr.Button("Submit")
-        clear = gr.Button("Clear Chat")
-
-    chatbot = gr.Chatbot(
-        label="Chat History"
-    )
-
-    # Initialize chat history as a Gradio state
-    chat_history = gr.State([])
-
-    def handle_submit(message, history, model, system_prompt, thinking_budget, api_key):
-        updated_history = process_chat(message, history, model, system_prompt, thinking_budget, api_key)
-        return updated_history, ""
-
-    def handle_clear():
-        return [], ""
-
-    submit.click(
-        handle_submit,
-        inputs=[msg, chat_history, model, system_prompt, thinking_budget, api_key],
-        outputs=[chatbot, msg]
-    )
-
-    clear.click(
-        handle_clear,
-        inputs=None,
-        outputs=[chatbot, msg]
-    )
-
-demo.launch()
+    chatbot = gr.Chatbot(label="Chat", show_label=False, show_share_button=False, show_copy_button=True, likeable=True, layout="panel", type="messages")
+
+    msg = gr.Textbox(label="Type your message here...", placeholder="Enter your message...")
+
+    gr.Button("Clear Chat").click(lambda: ([], ""), inputs=None, outputs=[chatbot, msg])
+
+    system_prompt = gr.Textbox(label="System Prompt", value=DEFAULT_SYSTEM_PROMPT, lines=15, interactive=True)
+
+    msg.submit(generate, inputs=[msg, chatbot, model, system_prompt, thinking_budget, api_key], outputs=[chatbot, msg])
+
+demo.launch(share=True, show_api=False) 

requirements.txt

@@ -1 +1 @@
-openai==0.28.0
+openai==1.45.1