Publications

One Model Definition, Many Accelerators: Scaling vLLM Across Hardware Without Forks