Large model weights look Gaussian, random Gaussian updates with learned scales work well for fine-tuning, and the paper builds simple adaptation and merging methods on that view.
Here, we also validate ∆W obtained from LoRA-Dash and DoRA
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unveiling the Mystery of Weight in Large Foundation Models: Gaussian Distribution Never Fades
Large model weights look Gaussian, random Gaussian updates with learned scales work well for fine-tuning, and the paper builds simple adaptation and merging methods on that view.