me-farza/lightreason: Local inference lab and interactive Studio for faster Qwen execution

Lightreason is a local inference lab and interactive Studio designed to run Qwen models faster on a single RTX 4090 GPU. The project implements techniques such as CUDA graphs, compact KV caches, and draft-assisted runners to accelerate Qwen2.5-3B-Instruct and Qwen3.6-35B-A3B. Benchmarks demonstrate substantial speed improvements across coding prompts while matching original output accuracy. Open article →