Luce-Org/lucebox

Why this repo matters

• •New LLM inference server with focus on speed and custom kernels.
• •Features optimizations like speculative prefill and expert offload.
• •Supports CUDA and HIP, indicating a focus on GPU acceleration.