week3/community-contributions/day5_openai_whisper_llamainstruct

import gradio as gr
import torch
from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig, TextStreamer, AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
from huggingface_hub import login
import os

# Use the secret stored in the Hugging Face space
token = os.getenv("HF_TOKEN")
login(token=token)

# Whisper Model Optimization
model = "openai/whisper-tiny"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

processor = AutoProcessor.from_pretrained(model)


transcriber = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    device=0 if torch.cuda.is_available() else "cpu",
)



# Function to Transcribe & Generate Minutes
def process_audio(audio_file):
    if audio_file is None:
        return "Error: No audio provided!"

    # Transcribe audio
    transcript = transcriber(audio_file)["text"]
    del transcriber 
    del processor
    # LLaMA Model Optimization
    LLAMA = "meta-llama/Llama-3.2-3B-Instruct"
    llama_quant_config = BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_use_double_quant=True,
        bnb_4bit_compute_dtype=torch.bfloat16,
        bnb_4bit_quant_type="nf4"
    )
    
    tokenizer = AutoTokenizer.from_pretrained(LLAMA)
    tokenizer.pad_token = tokenizer.eos_token
    model = AutoModelForCausalLM.from_pretrained(
        LLAMA,
        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
        device_map="auto"
    )
    # Generate meeting minutes
    system_message = "You are an assistant that produces minutes of meetings from transcripts, with summary, key discussion points, takeaways and action items with owners, in markdown."
    user_prompt = f"Below is an extract transcript of a Denver council meeting. Please write minutes in markdown, including a summary with attendees, location and date; discussion points; takeaways; and action items with owners.\n{transcript}"

    messages = [
        {"role": "system", "content": system_message},
        {"role": "user", "content": user_prompt}
    ]

    inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(DEVICE)
    streamer = TextStreamer(tokenizer)
    outputs = model.generate(inputs, max_new_tokens=2000, streamer=streamer)

    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Gradio Interface
interface = gr.Interface(
    fn=process_audio,
    inputs=gr.Audio(sources=["upload", "microphone"], type="filepath"),
    outputs="text",
    title="Meeting Minutes Generator",
    description="Upload or record an audio file to get structured meeting minutes in Markdown.",
)

# Launch App
interface.launch()