Fast LLM speculative inference server for consumer hardware.

An agent harness that compiles a model into one provably-correct, self-retargeting CUDA megakernel and self-tunes it past cuBLAS at batch-1 LLM decode.

Air.rs 70B+ inference on consumer GPU, LLM inference in Rust

A light, transparent, and modular inference & quantization engine for studying LLMs.

Qwen3-TTS inference with CUDA megakernels