Can SGlang load model weights from an existing instance with Ckpt Engine?

[XiaotaoChen]

I'm trying #11755 and #12216 to speed up loading model weights. And the workflow of init sglang instance for ckpt engine worker is clear. But i'm confused as to whether ckpt engine worker or SGLangCheckpointEngineWorkerExtension support init sglang instance from an existing instance? Seems Checkpoint Engine can support this scenario as it declare as belows. If support, what is the workflow like? I try it according the checkpoint engine doc, it raise error as below: torch.distributed.DistNetworkError: The server socket has failed to listen on any local network address. port: 29500, useIpv6: false, code: -98, name: EADDRINUSE, message: address already in use, detail info to see : MoonshotAI/checkpoint-engine#42

P2P: Used when new inference instances are dynamically added (due to restarts or dynamic availability) while the existing instances are already serving requests.

And i know --load-format remote_instance support this.

[stmatengss]

It is resolved.
MoonshotAI/checkpoint-engine#42