BERT: Sparsifying to Improve NLP Performance

Neural Magic creates models and recipes that allow anyone to plug in their data and leverage SparseML’s recipe-driven approach on top of Hugging Face’s robust training pipelines for the popular BERT NLP network. Sparsifying involves removing redundant information from neural networks using algorithms such as pruning and quantization, among others. This sparsification process results in faster inference and smaller file sizes for deployments.

This page walks through the following use cases for trying out the sparsified BERT models:

• Compare the differences between the models for both accuracy and inference performance

• Run the models for inference in deployment or applications

• Train the models on new datasets

Sparse Models

Card not loading? View card here.

Sparse Inference

Benchmark

Example Application

Example Deployment

Performance Comparisons

Coming soon!

Sparse Training

Sparse Transfer Learning

Coming soon!

Applying a Recipe

Creating a Recipe

Coming soon!