44 lines
1.3 KiB
Bash
44 lines
1.3 KiB
Bash
#!/bin/bash
|
|
|
|
# Download a test model for AMD GPU runner
|
|
echo "=== Downloading Test Model for AMD GPU ==="
|
|
echo ""
|
|
|
|
MODEL_DIR="models"
|
|
MODEL_FILE="$MODEL_DIR/current_model.gguf"
|
|
|
|
# Create directory if it doesn't exist
|
|
mkdir -p "$MODEL_DIR"
|
|
|
|
echo "Downloading SmolLM-135M (GGUF format)..."
|
|
echo "This is a small, fast model perfect for testing AMD GPU acceleration"
|
|
echo ""
|
|
|
|
# Download SmolLM GGUF model
|
|
wget -O "$MODEL_FILE" \
|
|
"https://huggingface.co/TheBloke/SmolLM-135M-GGUF/resolve/main/smollm-135m.Q4_K_M.gguf" \
|
|
--progress=bar
|
|
|
|
if [[ $? -eq 0 ]]; then
|
|
echo ""
|
|
echo "✅ Model downloaded successfully!"
|
|
echo "📁 Location: $MODEL_FILE"
|
|
echo "📊 Size: $(du -h "$MODEL_FILE" | cut -f1)"
|
|
echo ""
|
|
echo "🚀 Ready to start AMD GPU runner:"
|
|
echo "docker-compose up -d amd-model-runner"
|
|
echo ""
|
|
echo "🧪 Test the API:"
|
|
echo "curl http://localhost:11434/completion \\"
|
|
echo " -H 'Content-Type: application/json' \\"
|
|
echo " -d '{\"prompt\": \"Hello, how are you?\", \"n_predict\": 50}'"
|
|
else
|
|
echo ""
|
|
echo "❌ Download failed!"
|
|
echo "Try manually downloading a GGUF model from:"
|
|
echo "- https://huggingface.co/TheBloke"
|
|
echo "- https://huggingface.co/ggml-org/models"
|
|
echo ""
|
|
echo "Then place it at: $MODEL_FILE"
|
|
fi
|