Since posting this topic, I've set up the DS4 engine by Antirez on all of the GX10 (DGX Spark) machines I manage, to run Deepseek V4 Flash in parallel with LM Studio.

To serve multiple API endpoints (different applications/inference engines such as LM Studio, Unsloth, DS4, etc.) simultaneously on your own servers, just add another route to your Cloudflare networking config:

https://dash.cloudflare.com > Networking > Tunnels (select an exiting tunnel) > Routes > Add Route > Published Application

Make up a new subdomain name and enter the local network address:port of the application, as the Service URL in the Cloudflared route dialogue. For example, Antirez DS4 runs locally on http://localhost:8000, so that gets entered as the Service URL in the new Cloudflared route.

For each of the applications running on each of my servers, I have separate routes set up in each tunnel to each server where the local Cloudflared app is installed. For example, I followed the convention of naming the LM Studio route on each of my servers https://mysubdomain.mydomain.com and the DS4 route https://mysubdomainds4.mydomain.com

By following that pattern, you can set up as many application servers as needed, on each server machine you operate, each running on different ports on the same machine, by simply setting up separate Cloudflared routes to the local Service Address of each application (localhost:port).

I've been using this configuration in production since the beginning of July this year, and it's been 100% reliable.