Google builds a 600 billion parameter transformer to do massively multilingual, massive machine translation. Interestingly, the larger model scale does not come from increasing depth of the transformer, but from increasing width in the feedforward layers, combined with a hard routing to parallelize computations on up to 2048 TPUs. A very detailed engineering paper!
OUTLINE:
0:00 – Intro & Overview
4:10 – Main Results
5:10 – Mixture-of-Experts
16:00 – Difference to Scaling Classic Transformers
18:50 – Backpropagation in Mixture-of-Experts
20:05 – MoE Routing Algorithm in GShard
38:20 – GShard Einsum Examples
47:40 – Massively Multilingual Translation
56:00 – Results
1:11:30 – Conclusion & Comments
ERRATA:
I said the computation of MoE scales linearly, but actually, it’s sub(!)-linear.
Paper:
Abstract:
Neural network scaling has been critical for improving the model quality in many real-world machine learning applications with vast amounts of training data and compute. Although this trend of scaling is affirmed to be a sure-fire approach for better model quality, there are challenges on the path such as the computation cost, ease of programming, and efficient implementation on parallel devices. GShard is a module composed of a set of lightweight annotation APIs and an extension to the XLA compiler. It provides an elegant way to express a wide range of parallel computation patterns with minimal changes to the existing model code. GShard enabled us to scale up multilingual neural machine translation Transformer model with Sparsely-Gated Mixture-of-Experts beyond 600 billion parameters using automatic sharding. We demonstrate that such a giant model can efficiently be trained on 2048 TPU v3 accelerators in 4 days to achieve far superior quality for translation from 100 languages to English compared to the prior art.
Authors:
Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, Zhifeng Chen
Links:
YouTube:
Twitter:
Discord:
BitChute:
Minds:
source