Info

An FP8 scaled version of Chroma1 HD's latest, revised version (rev-0.1). Just used Clybius' conversion scripts from their GitHub repo to create these 2 versions:

  • Chroma1-HD_float8_e4m3fn_scaled_learned_svd
  • Chroma1-HD_float8_e4m3fn_scaled_learned_svd_fast

Note: I'm not sure why but HF has automatically categorized my version as a finetune for Chroma1 HD which it isn't. Just a scaled version.

More Info

I strongly suggest you download Clybius version if it's available somewhere in their HF repo. I'll actually remove those here once they upload their own fp8 versions, as I always use theirs. It just happened that I wanted to try this new revised Chroma1 HD version 0.1, and since Clybius hasn't uploaded their scaled version yet, I shamelessly took their conversion scripts and did it myself.

Needless to say all credit goes to lodestones for the amazing model and to Clybius for their scaled versions.

Some More Info

  • Do I have any clue whatsoever regarding optimal parameters or other refinements for the conversion process?

No, no I don't.

  • Can I guarantee I used the correct scripts and did the process as intended?

Nope, not that either.

I mean it's only a Python script for each version, so I guess I just run them and that's the intended process? Didn't get any errors during the process so that must be a plus too...

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for jtabox/Chroma1-HD_rev_v0.1_fp8-e4m3fn-scaled-learned-svd

Finetuned
(2)
this model