POET-X: Memory-efficient LLM Training by Scaling Orthogonal Transformation
This work addresses the memory and computational efficiency bottleneck in training large language models, making advanced training techniques more accessible for researchers and practitioners with limited hardware resources.
This paper introduces POET-X, a memory-efficient variant of the POET training framework for large language models (LLMs). POET-X significantly reduces computational cost and memory consumption, enabling the pretraining of billion-parameter LLMs on a single Nvidia H100 GPU, a task where standard optimizers like AdamW fail due to out-of-memory errors.
Efficient and stable training of large language models (LLMs) remains a core challenge in modern machine learning systems. To address this challenge, Reparameterized Orthogonal Equivalence Training (POET), a spectrum-preserving framework that optimizes each weight matrix through orthogonal equivalence transformation, has been proposed. Although POET provides strong training stability, its original implementation incurs high memory consumption and computational overhead due to intensive matrix multiplications. To overcome these limitations, we introduce POET-X, a scalable and memory-efficient variant that performs orthogonal equivalence transformations with significantly reduced computational cost. POET-X maintains the generalization and stability benefits of POET while achieving substantial improvements in throughput and memory efficiency. In our experiments, POET-X enables the pretraining of billion-parameter LLMs on a single Nvidia H100 GPU, and in contrast, standard optimizers such as AdamW run out of memory under the same settings.