Bag of Tricks for Optimizing Transformer Efficiency
This work provides practical, incremental improvements for researchers and practitioners seeking to optimize Transformer models without complex hardware dependencies.
The paper tackles the problem of improving Transformer efficiency by combining simple, hardware-agnostic methods like hyper-parameter tuning and better design choices, achieving speedups of 3.80X on CPU and 2.52X on GPU for inference on WMT news translation tasks.
Improving Transformer efficiency has become increasingly attractive recently. A wide range of methods has been proposed, e.g., pruning, quantization, new architectures and etc. But these methods are either sophisticated in implementation or dependent on hardware. In this paper, we show that the efficiency of Transformer can be improved by combining some simple and hardware-agnostic methods, including tuning hyper-parameters, better design choices and training strategies. On the WMT news translation tasks, we improve the inference efficiency of a strong Transformer system by 3.80X on CPU and 2.52X on GPU. The code is publicly available at https://github.com/Lollipop321/mini-decoder-network.