Unsloth Desktop: is image fine-tuning supported?

[mindplay-dk]

I wanted to try Unsloth Desktop to fine tune an image model, but I can't seem to find that feature?

The announcement on threads says it "runs and trains LLMs and diffusion models" - diffusion model could mean anything, I suppose, but it was mentioned next to LLMs, so I figured this meant image models?

I looked through the docs first and found this:

When I looked yesterday, "Fine-tune an image model" was a dead link - today it's not a link, but it still says "see", as though it was intended to be a link? But I can't find anything in the docs about fine tuning an image model - and in the app, there is just the one form called "Train" and the model choices seems to be filtered to text models only?

Pretty confusing. 😌

Is it planned and just not implemented or shipped yet, or what's going on?

[nicholas-greenwood]

try opening Images, then selecting Train at the top of that page. it's separate from the main Train page

the v0.1.908-beta source connects the Train tab to an image lora training panel

[mindplay-dk]

Interesting choice - not where I expected to find that, and for some reason I couldn't see it at all. 😅

This might be clearer as a separate item in the main nav? Not sure. Training doesn't seem like just a sub feature of image generation? Text training gets a top level "Train" entry in the main nav - and this definitely seems like a main feature? 😊