app.py

import gradio as gr
import torch
from transformers import pipeline, WhisperProcessor, WhisperForConditionalGeneration, AutoModelForCausalLM, AutoTokenizer
from huggingface_hub import login
import os

# Retrieve the token from the environment variable
hf_api_token = os.getenv("HF_API_TOKEN")

if hf_api_token is None:
    raise ValueError("HF_API_TOKEN environment variable is not set")

# Authenticate with Hugging Face
login(token=hf_api_token, add_to_git_credential=True)

# Initialize the Whisper processor and model
whisper_processor = WhisperProcessor.from_pretrained("openai/whisper-base")
whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")

# Initialize the summarization model and tokenizer
# Use a smaller version of the Llama model and load in FP16
summarization_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/LlamaGuard-7b",
    torch_dtype=torch.float16,
    low_cpu_mem_usage=True
)
summarization_tokenizer = AutoTokenizer.from_pretrained("meta-llama/LlamaGuard-7b")

# Function to transcribe audio
def transcribe_audio(audio_file):
    # Load audio file
    audio_input, _ = whisper_processor(audio_file, return_tensors="pt", sampling_rate=16000).input_values
    # Generate transcription
    transcription_ids = whisper_model.generate(audio_input)
    transcription = whisper_processor.decode(transcription_ids[0])
    return transcription

# Function to summarize text
def summarize_text(text):
    inputs = summarization_tokenizer(text, return_tensors="pt", max_length=512, truncation=True)
    summary_ids = summarization_model.generate(inputs.input_ids, max_length=150, min_length=40, length_penalty=2.0, num_beams=4, early_stopping=True)
    summary = summarization_tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return summary

# Gradio interface
def process_audio(audio_file):
    transcription = transcribe_audio(audio_file)
    summary = summarize_text(transcription)
    return transcription, summary

# Gradio UI
iface = gr.Interface(
    fn=process_audio,
    inputs=gr.Audio(source="upload", type="file"),
    outputs=[
        gr.Textbox(label="Transcription"),
        gr.Textbox(label="Summary")
    ],
    title="Audio Transcription and Summarization",
    description="Upload an audio file to transcribe and summarize the conversation."
)

# Launch the app
iface.launch()