ekzhang / resin

Why you might care

Aimed at developers who want to hack on inference systems or new models and hardware, not run in production. You supply model-specific kernels; Resin provides the Rust batch scheduler, slab-based cache pool, CUDA graphs, parallelism, and optional HTTP serving with Prometheus metrics. The README warns it is not production-quality and points to SGLang or vLLM for production use.