setup_huggingface.sh

#!/bin/bash

# Make script exit on error
set -e

echo "Setting up LLaMA-Omni on Hugging Face Spaces..."

# Downgrade pip to avoid OmegaConf metadata validation issues
echo "Downgrading pip to version 24.0 to avoid OmegaConf metadata validation issues..."
python -m pip install --upgrade "pip<24.1"

# Create necessary directories
mkdir -p models/speech_encoder vocoder

# Download vocoder models if needed
if [ ! -f "vocoder/g_00500000" ]; then
    echo "Downloading vocoder models..."
    wget -P vocoder/ https://dl.fbaipublicfiles.com/fairseq/speech_to_speech/vocoder/code_hifigan/mhubert_vp_en_es_fr_it3_400k_layer11_km1000_lj/g_00500000
    wget -P vocoder/ https://dl.fbaipublicfiles.com/fairseq/speech_to_speech/vocoder/code_hifigan/mhubert_vp_en_es_fr_it3_400k_layer11_km1000_lj/config.json
fi

# Create empty __init__.py files for the package structure
mkdir -p omni_speech/serve omni_speech/infer/examples
touch omni_speech/__init__.py
touch omni_speech/serve/__init__.py
touch omni_speech/infer/__init__.py

echo "✅ Setup complete! LLaMA-Omni is now ready to run on Hugging Face Spaces."