Tricks for Training Sparse Translation Models
This work addresses training challenges in sparse multilingual translation models, which is an incremental improvement for machine translation efficiency.
The paper tackles the problem of poor performance in sparse architectures for multilingual machine translation by proposing two techniques: temperature heating and dense pre-training. These methods improve performance on two benchmarks and more than double model convergence speed when combined.
Multi-task learning with an unbalanced data distribution skews model learning towards high resource tasks, especially when model capacity is fixed and fully shared across all tasks. Sparse scaling architectures, such as BASELayers, provide flexible mechanisms for different tasks to have a variable number of parameters, which can be useful to counterbalance skewed data distributions. We find that that sparse architectures for multilingual machine translation can perform poorly out of the box, and propose two straightforward techniques to mitigate this - a temperature heating mechanism and dense pre-training. Overall, these methods improve performance on two multilingual translation benchmarks compared to standard BASELayers and Dense scaling baselines, and in combination, more than 2x model convergence speed.