49What is model merging (e.g. model soups, task arithmetic), and why is it useful?▼hardHugging FaceGoogleMistral1 replies◆ premiumFuse several fine-tunes into one model by doing arithmetic on their weights, with no retraining and no data. The signal is knowing why it works and which method to pick when merges interfere.Open full answer →
122How do you merge or compose multiple LoRA adapters, and when does it break?▼hardHugging FacePredibaseNVIDIA◆ premiumMerging one adapter is a single matrix add. Merging four is where careers get interesting: the low-rank updates were each trained as if the others did not exist, and they collide. Here is what actually works and when to route instead.Open full answer →
