Clarification for Incremental Weights Only
#75
by boyden - opened
For reproducibility, we would like to clarify the following: Can we open-source ONLY the incremental weights (such as LoRA adapters or full-parameter delta weights $A-B$)? Our code will strictly use timm.create_model("hf_hub:paige-ai/Virchow2") as the first step, forcing all downstream users to authenticate via their own Hugging Face accounts to access the base model. Does distributing only the increments violate the "ND" or "not to distribute" clauses?