SkyRL brings Tinker to your GPUs (2025)

We are excited about the Tinker API, it exposes the primitives forward_backward, optim_step, sample and checkpoints of an LLM as a REST API, which can be used to implement pretty much arbitrary training recipes while hiding all the infra challenges of running the model and also abstract the underlying accelerator with very small surface area. We hope it can emerge as a standard. If you have any feedback for our open-source implementation we would love to hear about it!

The date is wrong. It is 2026, not 2025