[FEATURE Request] - DRuGS sampler: Deep Random Micro-Glitch Sampling

[SabinStargem]

It is a sampler that injects noise into layers of the AI, changing the output. Apparently the AI can overcome this noise, but the results will be mildly distorted from doing so. Not being a smartie, I recommend using the link for actual details. Much like XTC, it holds promise for increasing the creativity of a model.

DRuGS Github

[LostRuins]

That modifies the model weights itself, so I think it might not be ideal for KCPP. I am looking into implementing XTC first.

[envy-ai]

"DRµGS just inverts this scheme. Instead of using noise to sample from the model's predictions, DRµGS injects noise directly into the transformer layers at inference time, thereby varying what the model predicts. From here, simply selecting the most likely prediction is often enough to increase output variety while maintaining coherence."

Looks like it's happening during inference as opposed to as a modification to the model file.