Jianfei Chen

A new approach to reduce memory use in neural networks through 4-bit optimizers.

VCAS improves neural network training efficiency without losing accuracy.

A new method enhances sparse language model training while minimizing performance loss.

Introducing S-STE, a novel approach to improve sparse neural network training efficiency.

A new method speeds up AI processing without losing accuracy.

ReMoE brings flexibility and efficiency to language models with dynamic expert selection.