This page provides an informal, technically-oriented survey over available (and commonly used) architectures and implementations for large-scale pre-training (and fine-tuning) of contextualized neural language models.
The NLPL use case, will install, validate, and maintain a selection of these implementations, in an automated and uniform manner, on multiple HPC systems.
Bidirectional Encoder Representations from Transformers (BERT) is a deep language model jointly conditioned on both left and right context in all layers. It is based on the Transformer neural architecture (Devlin et al 2019).
The de-facto standard for contextualized representations in modern NLP.
- Reference Google implementation in TensorFlow. Requirements: 1.11 <= TensorFlow < 2.0.
- HuggingFace Transformers implementation. Can train either with TensorFlow or with PyTorch. Requirements: Python >=3.6, TensorFlow >= 2.0, PyTorch >=1.3.1.
- Chainer implementation. Not much interesting to us, since it does not support training, only inference.
Embeddings from Language Models (ELMo) use bidirectional LSTM language models to produce contextualized word token representations (Peters et al 2018).
The only architecture in the list to use recurrent neural networks, not Transformers. Despite being much less computationally demanding, often performs on par with BERT.
- Reference Tensorflow implementation. Requirements: Python >=3.5, 1.2 < TensorFlow < 1.13 (later versions produce too many deprecation warnings), h5py.
- LTG implementation. Based on the reference implementation, but with improved data loading, hyperparameter handling, and the code updated to more recent versions of TensorFlow. Requirements: Python >=3.5, 1.15 <= Tensorflow < 2.0 (2.0 version is planned), h5py, smart_open. Tutorial is available.
- PyTorch implementation in AllenNLP. Not much interesting to us, since it does not support training, only inference. Requirements: Python >= 3.6, 1.6 <= PyTorch < 1.7.
Robustly Optimized BERT (RoBERTa) is a BERT variation by Facebook. The most important changes are removing the next sentence prediction objective and dynamically changing the masking pattern applied to the training data. Otherwise, it is just BERT on steroids (training longer, bigger batches, longer sequences). Interestingly, the RoBERTa paper was rejected by ICLR 2020.