scripts/setup_caption.sh

#!/bin/bash
# Dataset-Prep v3 — Caption setup
# Includes ALL base deps + Ollama + Qwen2.5-VL captioning model
set -e

echo "=== Dataset-Prep v3: Caption Setup ==="

# Run base setup first
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
if [ -f "$SCRIPT_DIR/setup_base.sh" ]; then
    bash "$SCRIPT_DIR/setup_base.sh"
else
    python3 -c "
from huggingface_hub import hf_hub_download
path = hf_hub_download('msrcam/ds-prep-backend', 'scripts/setup_base.sh', repo_type='dataset')
print(path)
" > /tmp/base_path.txt
    bash "$(cat /tmp/base_path.txt)"
fi

# Install Ollama
echo "Installing Ollama..."
if command -v ollama &> /dev/null; then
    echo "Ollama already installed."
else
    curl -fsSL https://ollama.com/install.sh | sh
fi

# Start Ollama
export OLLAMA_NUM_PARALLEL=4
nohup ollama serve > /tmp/ollama.log 2>&1 &
echo $! > /tmp/ollama.pid
sleep 5

# Pull caption model
echo "Pulling Qwen2.5-VL 7B..."
ollama pull huihui_ai/qwen2.5-vl-abliterated:7b

echo "=== Caption setup complete ==="