pixi run max serve \
  --model openai-community/gpt2 \
  --custom-architectures gpt2_module_v3 \
  --port 8888

curl http://localhost:8888/v1/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai-community/gpt2",
    "prompt": "The future of AI",
    "max_tokens": 30
  }' | jq .