# A new voice without touching every weight

> Low-rank adapters: train under 1% of the numbers and keep the same speed.

PyTorch · November 9, 2026 · Pedro Cormann, Founder & Machine Learning Engineer · https://lero.studio/research/en/baixo-rank

A linear layer stores a d × d matrix W. In full training, every number changes. With a low-rank adapter, W stays frozen and only two thin matrices are trained, B (d × r) and A (r × d). The update is the product B·A, added to W with weight α/r. With d = 4096 and r = 16, that is 131 thousand trainable numbers instead of 16.7 million: under 1%.

_(Interactive on the page: https://lero.studio/research/en/baixo-rank)_

Training gets cheap and fast, and each voice becomes a small file. Later, B·A can be merged into W, so inference is as fast as the original model.

_(Interactive on the page: https://lero.studio/research/en/baixo-rank)_