Load multiple DIT models from single server?

[theaerotoad]

Now that the server can (lazy) unload models when not in use, would you consider having the server offer multiple DIT options? I think there's a way to do that with the acestep python via environment variables (e.g. ACESTEP_CONFIG_PATH, ACESTEP_CONFIG_PATH2, up to three). Might be as easy as just using --dit multiple times on the CLI, with default being either the first or last one?

It might be nice to have up to three, as Turbo, SFT, and Base are have useful attributes. Maybe a drop down to pick the model on the webui?

ps. The webui is coming along nicely, too!

[ServeurpersoCom]

:D I just saw this issue and I've already coded this:

Put your GGUF inside models/ (any LM, DiT you want, and the Qwen3 text-encoder + VAE)
Put your LoRAs inside loras/ (ComfyUI .safetensors or PEFT subdirectory)

The API is very simple and parses the folder. In the request, you choose the model you want. The interface remembers them, but the JSON also remembers the used model, and automatically swaps it

Done :D

[ServeurpersoCom]

Now I separate the technical documentation from the user-friendly documentation, I add a binary link, and voilà, on Windows, a few downloads, one click on server.sh or server.cmd, navigate to 127.0.0.1:8085 and go