Snapshot gets GPU pods ready in seconds instead of minutes — by restoring a fully initialized GPU worker instead of starting one from scratch. Kubernetes-native checkpoint & restore that runs alongside your existing stack.
GPU checkpoint/restore for vLLM: snapshot a warmed-up engine (compiled kernels, CUDA graphs, weights, 1-8 GPUs) into an OCI image and restore it in seconds instead of a multi-minute cold start. Scale to zero, spot GPUs, Kubernetes optional.
Serverless-GPU LLM serving: scale-to-zero with fast GPU snapshot/restore (cuda-checkpoint), multi-tenant packing, and an OpenAI-compatible API — built on vLLM.
A job queue for the GPU that runs your local model: training, evals and benchmarks run in order, beside the model or in its idle gaps, and step aside when a request arrives.
To associate your repository with the cuda-checkpoint topic, visit your repo's landing page and select "manage topics."