Summary of Llm Pruning and Distillation in Practice: the Minitron Approach, by Sharath Turuvekere Sreenivas et al.

LLM Pruning and Distillation in Practice: The Minitron Approach

by Sharath Turuvekere Sreenivas, Saurav Muralidharan, Raviraj Joshi, Marcin Chochowski, Ameya Sunil Mahabaleshwarkar, Gerald Shen, Jiaqi Zeng, Zijia Chen, Yoshi Suhara, Shizhe Diao, Chenhan Yu, Wei-Chun Chen, Hayley Ross, Oluwatobi Olabiyi, Ashwath Aithal, Oleksii Kuchaiev, Daniel Korzekwa, Pavlo Molchanov, Mostofa Patwary, Mohammad Shoeybi, Jan Kautz, Bryan Catanzaro

First submitted to arxiv on: 21 Aug 2024

GrooveSquid.com Paper Summaries

GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!

Summary difficulty	Written by	Summary
High	Paper authors	High Difficulty Summary Read the original abstract here
Medium	GrooveSquid.com (original content)	Medium Difficulty Summary This comprehensive report explores compressing large language models using pruning and distillation techniques. The authors focus on reducing the parameters of Llama 3.1’s 8B and Mistral NeMo’s 12B models to 4B and 8B, respectively. Two pruning strategies are employed: depth pruning and joint hidden/attention/MLP pruning, which are evaluated on common benchmarks from the LM Evaluation Harness. The pruned models are then fine-tuned using instruct-tuning and tested with the NeMo Aligner. The results show that a compelling 4B model can be derived from Llama 3.1’s 8B and a state-of-the-art MN-Minitron-8B model from Mistral NeMo’s 12B.
Low	GrooveSquid.com (original content)	Low Difficulty Summary This study shows how to make large language models smaller without losing their power. The researchers used special techniques called pruning and distillation to shrink the size of two big models, Llama 3.1 and Mistral NeMo. They tried different ways of pruning and found that a combination of methods worked best. The smaller models were then tested on some common benchmarks and showed impressive results.