Optimization and Computational Efficiency

Low-Rank Inference

Approach that approximates the model's large weight matrices by products of lower-rank matrices, drastically reducing the number of parameters and matrix multiplication operations during inference.

← Wstecz