BLoad: Enhancing Neural Network Training with Efficient Sequential Data Handling
This addresses a scalability problem for researchers and practitioners dealing with large, variable-length sequential datasets, though it appears incremental as it builds on existing distributed data-parallel methods.
The paper tackled the challenge of efficiently training neural networks on sequences of varying sizes by proposing a novel training scheme that reduces padding by over 100x without deleting frames, resulting in improved training time and Recall in experiments.
The increasing complexity of modern deep neural network models and the expanding sizes of datasets necessitate the development of optimized and scalable training methods. In this white paper, we addressed the challenge of efficiently training neural network models using sequences of varying sizes. To address this challenge, we propose a novel training scheme that enables efficient distributed data-parallel training on sequences of different sizes with minimal overhead. By using this scheme we were able to reduce the padding amount by more than 100$x$ while not deleting a single frame, resulting in an overall increased performance on both training time and Recall in our experiments.