What exactly is this?
i noticed you mixed both Magmell and VioletGRPO BACK into MagCap... why? What exactly do you hope to achieve? Im curious of your thought process.
i noticed you mixed both Magmell and VioletGRPO BACK into MagCap... why? What exactly do you hope to achieve? Im curious of your thought process.
This was originally meant to serve as a foundation for another series of reasoning runs, though I can’t recall why I ultimately didn’t use it. The core idea was to prevent catastrophic forgetting and help rebalance the model series—especially since most of the continued training relied on QLoRA adapters. As a general rule with models in this archive organization: they’re typically either experimental or works in progress. I wouldn’t recommend trying to use anything that's been relegated here.