The release of the training code

[Sainzerjj]

May I ask whether there are plans to open-source the training code in the future, and if so, approximately how long would that take? Thank you very much.

[Dervlex]

@Sainzerjj Training code?
I am not OP, but isnt this just a:
User prompt -> LLM refining prompt -> KSampler (used refined prompt instead of user prompt)

Inference ?
And nothing special?

Look into inference.py

[Dervlex]

@Sainzerjj ahhhh i see. They implemented also the Dual-GRPO 👍

[Dervlex]

But i cant find it inside the "inference.py" 🤔🤔🤔🤔🤔🤔

[Sainzerjj]

They provided the weights after SFT + dual GRPO training. The inference code, apart from LLM refining, indeed has no difference from normal inference.