start.sh

#!/bin/bash
set -e

echo "================================================"
echo "🚀 Axon - Qwen2.5-Omni-7B Multimodal Server"
echo "================================================"
echo ""
echo "📋 Capabilities: Text | Images | Audio"
echo "🔧 Quantization: Q8_0 (near-lossless)"
echo "⚡ Optimizations: Flash Attention, Continuous Batching"
echo ""

mkdir -p /app/models/qwen2.5-omni-7b

download_file () {
    REPO_ID=$1
    FILENAME=$2
    DEST_PATH=$3

    if [ -f "$DEST_PATH" ]; then
        echo "✅ Already exists: $(basename $DEST_PATH)"
        return 0
    fi

    echo "⬇️  Downloading $FILENAME ..."
    python3 -c "
from huggingface_hub import hf_hub_download
import shutil, sys
try:
    path = hf_hub_download(
        repo_id='$REPO_ID',
        filename='$FILENAME',
        cache_dir='/app/.cache'
    )
    shutil.copy(path, '$DEST_PATH')
    print(f'✅ Downloaded: $FILENAME')
except Exception as e:
    print(f'❌ Error downloading $FILENAME: {e}')
    sys.exit(1)
"
}

download_file "ggml-org/Qwen2.5-Omni-7B-GGUF" \
    "Qwen2.5-Omni-7B-Q8_0.gguf" \
    "/app/models/qwen2.5-omni-7b/Qwen2.5-Omni-7B-Q8_0.gguf"

download_file "ggml-org/Qwen2.5-Omni-7B-GGUF" \
    "mmproj-Qwen2.5-Omni-7B-Q8_0.gguf" \
    "/app/models/qwen2.5-omni-7b/mmproj-Qwen2.5-Omni-7B-Q8_0.gguf"

echo ""
echo "🚀 Starting llama.cpp Server"
echo "🌐 Server will be available at http://0.0.0.0:7860"
echo ""

exec /usr/local/bin/llama-server \
    --models-dir /app/models \
    --host 0.0.0.0 \
    --port 7860 \
    -c 8192 \
    -t 4 \
    -fa on \
    -cb \
    --n-gpu-layers 0