Reparameterized LLM Training via Orthogonal Equivalence Transformation
This addresses training stability and efficiency issues for LLM developers, though it appears incremental as it builds on existing reparameterization techniques.
The paper tackles the challenge of effectively and reliably training large language models (LLMs) by proposing POET, a reparameterized training algorithm using orthogonal equivalence transformation, which improves generalization and scalability in experiments.
While large language models (LLMs) are driving the rapid advancement of artificial intelligence, effectively and reliably training these large models remains one of the field's most significant challenges. To address this challenge, we propose POET, a novel reParameterized training algorithm that uses Orthogonal Equivalence Transformation to optimize neurons. Specifically, POET reparameterizes each neuron with two learnable orthogonal matrices and a fixed random weight matrix. Because of its provable preservation of spectral properties of weight matrices, POET can stably optimize the objective function with improved generalization. We further develop efficient approximations that make POET flexible and scalable for training large-scale neural networks. Extensive experiments validate the effectiveness and scalability of POET in training LLMs.