Bert Squeeze: compress Transformer models

Bert Squeeze

Bert Squeeze is a Python library of code for reducing the size or inference latency of Transformer-based models, with implementations for distillation, pruning and quantization and early exiting listed among its techniques.

Independent OpenViST project profile, based on linked sources.

Understand the project

Practical overview

When it fits

Use it when developing a Transformer model workflow in PyTorch Lightning and you want code and examples for model compression techniques, especially for sequence classification.

Who it is for

Developers training or optimizing Transformer-based sequence-classification models.
Practitioners exploring model compression before deploying Transformer components.

Documented uses

Train a smaller model to mimic a larger teacher model through distillation.
Reduce model weights, layers or attention heads through pruning.
Apply dynamic quantization or quantization-aware training.
Explore early exiting as an inference-latency approach.

Before you choose

Requirements & limits

The README identifies PyTorch Lightning and Transformers as the project’s framework ecosystem.
The stated current model-task scope is sequence classification.
The repository cautions that not all procedures have been extensively tested and makes no performance guarantee for every method.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up the Bert Squeeze repository

Download the repository and install its dependencies in a uv-managed environment.

Before you start

Python project environment with uv available
Access to the Bert Squeeze GitHub repository

Steps

Clone https://github.com/JulesBelveze/bert-squeeze.git.
Create a virtual environment with the command below.
Documented command
```
uv venv
```
Activate it with source .venv/bin/activate.
Install dependencies with the command below.
Documented command
```
uv sync
```

Expected outcome

The repository dependencies are installed in the activated environment.

Guide limitations

This setup procedure does not describe running a training or compression workflow.
The author says the procedures have not all been extensively tested and does not guarantee every method’s performance.

Documentation 1 ↗

Evidence and transparency

Sources & review

Official repository ↗

Source record details

Source availability: Active
Sources checked: 18 October 2026
Original website title: JulesBelveze/bert-squeeze · Official homepage title · retained in its source language

How we review sources Suggest a correction