| This is a 800M parameters model pre-trained with [QuEST](https://arxiv.org/abs/2502.05003) over 80B C4 tokens in 2:4 sparse INT4 format. | |
| The code to verify that this model works in INT4 can be found [here](https://github.com/IST-DASLab/QuEST/blob/main/src/HadamardFourEightTesting.ipynb). |