SpotWarp, the GPU eviction escape hatch

Field notes

The project presents itself as a local guard process rather than a cluster orchestrator: users wrap an existing command, while the daemon is intended to stream delta checkpoint slices and recover on another GPU provider after eviction. Its stated target is the marketplace layer around Vast.ai and RunPod, with examples including RTX 4090 instances at $0.34 per hour and A100 instances at $0.80 per hour. The page also lists PyTorch, Hugging Face, ComfyUI, and vLLM integrations, though the operational claims are not independently verified here.