Update app.py

app.py

@@ -1,118 +1,64 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
 import random
-import textwrap
-from transformers import pipeline
-import numpy as np
 
-# Load the Whisper model for automatic speech recognition
-transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-base.en")
-
-# Define the model to be used
+# Initialize the model
 model = "mistralai/Mixtral-8x7B-Instruct-v0.1"
 client = InferenceClient(model)
 
-# Embedded system prompt
-system_prompt_text = (
-    "You are a smart and helpful co-worker of Thailand based multi-national company PTT, and PTTEP. "
-    "You help with any kind of request and provide a detailed answer to the question. But if you are asked about something "
-    "unethical or dangerous, you must refuse and provide a safe and respectful way to handle that."
-)
-
-# Function to transcribe audio input
-def transcribe(audio):
-    if audio is None:
-        return None  # Handle case where audio input is None
-
-    sr, y = audio
-    # Convert to mono if stereo
-    if y.ndim > 1:
-        y = y.mean(axis=1)
-
-    y = y.astype(np.float32)
-    y /= np.max(np.abs(y))  # Normalize audio
-
-    return transcriber({"sampling_rate": sr, "raw": y})["text"]  # Transcribe audio
-
-def format_prompt_mixtral(message, history):
-    prompt = "<s>"
-    prompt += f"{system_prompt_text}\n\n"  # Add the system prompt
-
-    if history:
-        for user_prompt, bot_response in history:
-            prompt += f"[INST] {user_prompt} [/INST]"
-            prompt += f" {bot_response}</s> "
-    prompt += f"[INST] {message} [/INST]"
-    return prompt
-
-def chat_inf(prompt, history, seed, temp, tokens, top_p, rep_p):
-    generate_kwargs = dict(
-        temperature=temp,
-        max_new_tokens=tokens,
-        top_p=top_p,
-        repetition_penalty=rep_p,
-        do_sample=True,
-        seed=seed,
-    )
-
-    formatted_prompt = format_prompt_mixtral(prompt, history)
-    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+def chat_response(prompt, history, seed, temp, tokens, top_p, rep_p):
+    generate_kwargs = {
+        "temperature": temp,
+        "max_new_tokens": tokens,
+        "top_p": top_p,
+        "repetition_penalty": rep_p,
+        "do_sample": True,
+        "seed": seed,
+    }
+
+    # Include the chat history in the prompt
+    formatted_prompt = "\n".join([f"Q: {user_prompt}\nA: {bot_response}" for user_prompt, bot_response in history]) + f"\nQ: {prompt}\nA:"
+    
     output = ""
-    for response in stream:
-        output += response.token.text
-        yield [(prompt, output)]
-    history.append((prompt, output))
-    yield history
+    
+    # Generating text in streaming mode
+    for response in client.text_generation(formatted_prompt, **generate_kwargs, stream=True):
+        # Assuming response is directly a string or contains a message
+        output += response  # Using response directly since it's a string
 
-def clear_fn():
-    return None, None
+        # Yield the updated output for real-time display
+        yield [(prompt, output)]
 
-rand_val = random.randint(1, 1111111111111111)
+    # Append the full response to history after completion
+    history.append((prompt, output))
+    yield history  # Yielding the updated history
 
-def check_rand(inp, val):
-    if inp:
-        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1, 1111111111111111))
-    else:
-        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=int(val))
+def clear_chat():
+    return [], []  # Returning an empty history
 
+# Gradio interface
 with gr.Blocks() as app:
-    gr.HTML("""<center><h1 style='font-size:xx-large;'>PTT Chatbot</h1><br><h3>running on Huggingface Inference</h3><br><h7>EXPERIMENTAL</center>""")
+    gr.HTML("<center><h1>Chatbot</h1><h3>Ask your questions!</h3></center>")
     
-    with gr.Row():
-        chat = gr.Chatbot(height=500)
-
-    with gr.Group():
-        with gr.Row():
-            with gr.Column(scale=3):
-                inp = gr.Audio(type="filepath")  # Remove the source parameter
-                with gr.Row():
-                    with gr.Column(scale=2):
-                        btn = gr.Button("Chat")
-                    with gr.Column(scale=1):
-                        with gr.Group():
-                            stop_btn = gr.Button("Stop")
-                            clear_btn = gr.Button("Clear")
-            with gr.Column(scale=1):
-                with gr.Group():
-                    rand = gr.Checkbox(label="Random Seed", value=True)
-                    seed = gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, step=1, value=rand_val)
-                    tokens = gr.Slider(label="Max new tokens", value=3840, minimum=0, maximum=8000, step=64, interactive=True, visible=True, info="The maximum number of tokens")
-                    temp = gr.Slider(label="Temperature", step=0.01, minimum=0.01, maximum=1.0, value=0.9)
-                    top_p = gr.Slider(label="Top-P", step=0.01, minimum=0.01, maximum=1.0, value=0.9)
-                    rep_p = gr.Slider(label="Repetition Penalty", step=0.1, minimum=0.1, maximum=2.0, value=1.0)
-
-    def handle_chat(audio_input, chat_history, seed, temp, tokens, top_p, rep_p):
-        user_message = transcribe(audio_input)  # Transcribe audio to text
-        if user_message is None or user_message == "":  # Check for empty or error in recognition
-            return chat_history, "Sorry, I couldn't understand that."
-
-        response_gen = chat_inf(user_message, chat_history, seed, temp, tokens, top_p, rep_p)
-        response = next(response_gen)[0][-1][1]  # Get the response text
-        return chat_history + [(user_message, response)], response  # Return updated chat history
-
-    go = btn.click(handle_chat, [inp, chat, seed, temp, tokens, top_p, rep_p], chat)
+    chat_box = gr.Chatbot(height=500)
+    inp = gr.Textbox(label="Your Question", lines=5)
+    btn = gr.Button("Ask")
+    clear_btn = gr.Button("Clear")
+    
+    rand_seed = gr.Checkbox(label="Random Seed", value=True)
+    seed_slider = gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1, 1111111111111111))
+    tokens_slider = gr.Slider(label="Max new tokens", value=3840, minimum=0, maximum=8000)
+    temp_slider = gr.Slider(label="Temperature", value=0.9, minimum=0.01, maximum=1.0)
+    top_p_slider = gr.Slider(label="Top-P", value=0.9, minimum=0.01, maximum=1.0)
+    rep_p_slider = gr.Slider(label="Repetition Penalty", value=1.0, minimum=0.1, maximum=2.0)
+
+    # Handle button click to get chat response
+    btn.click(
+        lambda prompt: chat_response(prompt, [], seed_slider.value, temp_slider.value, tokens_slider.value, top_p_slider.value, rep_p_slider.value),
+        inp,
+        chat_box,
+    )
 
-    stop_btn.click(None, None, None, cancels=[go])
-    clear_btn.click(clear_fn, None, [inp, chat])
+    clear_btn.click(clear_chat, None, [inp, chat_box])
 
-app.queue(default_concurrency_limit=10).launch(share=True, auth=("admin", "0112358"))  # Launch the app with authentication
+app.launch(share=True, auth=("admin", "0112358"))