Integrate CSGHub with dstack for apps, training, and inference

[peterschmidt85]

CSGHub integrates with Kubernetes, Ray, and Knative to deploy apps, train models, and run model inference.

dstack is a lightweight alternative to Kubernetes, Ray, and Knative that could be a much better fit.

Here are its key advantages:

1. dstack supports leading cloud GPU providers and makes it easy to add native support for new providers.
2. dstack simplifies the use of on-prem GPU machines compared to Kubernetes.
3. dstack addresses many challenges related to deploying apps, managing endpoints, and scheduling complex training tasks.

By integrating with dstack, CSGHub users could experience up to a 10x simplification when connecting to cloud providers or on-prem clusters.

[peterschmidt85]

If required, our team is happy to assist with the integration.

[ganisback]

Thank you for reporting feature requirement, I just looked at it, dstack is a very light framework for LLM deployment. we will start investigation in next a few weeks.

[samchen8008]

Guys, @peterschmidt85 @ganisback can you guys create one PR for the further integration?