Pytorch

[Notes] Gradient Checkpointing with BERT

A brief analysis of huggingface’s implementation

[Paper] Adafactor: Adaptive Learning Rates with Sublinear Memory Cost

Essential for fine-tuning T5…

[Paper] Rethinking Cooperative Rationalization: Introspective Extraction and Complement Control

[PyTorch Lightning] Log Training Losses when Accumulating Gradients

The global step is not what you…

Generating Synthetic Tabular Data Using GAN

A case study: detecting credit fraud

[Paper] Language-agnostic BERT Sentence Embedding

Improving accuracy for low-resource languages

[Paper] Training Question Answering Models From Synthetic Data

Generate questions and answers from…

[Tip] TorchScript Supports Half Precision

Speeding up inference for models trained with mixed…

Deploying EfficientNet Model using TorchServe

A Case Study

Fine-tuning BERT for Similarity Search

BERT as Sentence Encoder is Surprisingly Sample-Efficient