We present a comprehensive report on compressing the Llama 3.1 8B and Mistral NeMo 12B models to 4B and 8B parameters, respectively, using pruning and distillation. We explore two distinct pruning strategies: (1) depth pruning and (2) joint hidden/attention/MLP (width) pruning, and evaluate the results on common benchmarks from the LM Evaluation Harness. The models are then aligned with NeMo Aligner and tested in instruct-tuned versions. This approach produces a compelling 4B model from Llama 3.1 8B and a state-of-the-art Mistral-NeMo-Minitron-8B (MN-Minitron-8B for brevity) model from Mistral NeMo 12B. We found that with no access to the original data, it is beneficial to slightly fine-tune teacher models on the distillation dataset. We open-source our base model weights on Hugging Face with a permissive license.
Cite
@article{arxiv.2408.11796,
title = {LLM Pruning and Distillation in Practice: The Minitron Approach},
author = {Sharath Turuvekere Sreenivas and Saurav Muralidharan and Raviraj Joshi and Marcin Chochowski and Ameya Sunil Mahabaleshwarkar and Gerald Shen and Jiaqi Zeng and Zijia Chen and Yoshi Suhara and Shizhe Diao and Chenhan Yu and Wei-Chun Chen and Hayley Ross and Oluwatobi Olabiyi and Ashwath Aithal and Oleksii Kuchaiev and Daniel Korzekwa and Pavlo Molchanov and Mostofa Patwary and Mohammad Shoeybi and Jan Kautz and Bryan Catanzaro},
journal= {arXiv preprint arXiv:2408.11796},
year = {2024}
}