TensorRails/warmctl
在 GitHub 打开 ↗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.
Star30
24h 新增—
7 天新增—
Fork0
Issue12
语言Go
项目简介
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.
Star 趋势
话题标签
checkpoint-restorecold-startcriucuda-checkpointgpukubernetesllm-inferencellm-servingnvidiaociscale-to-zeroserverlessvllm
仓库信息
| License | Apache-2.0 |
| 创建时间 | 2026-10-12 |
| 最后推送 | 1 天前 |
| 热度分 | 0.0 |