Song Han
Professor · MIT · United States
Led model compression from Deep Compression to AWQ, the activation-aware weight quantization used to run LLMs on small hardware.
- Areas
- Quantization, Efficiency
- Eras
- Efficiency, Scaling
- Top era
- Efficiency
- Top score
- 8
- Confidence
- high
Key publications
- Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding (2016, ICLR)
- AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration (2024, MLSys)