{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/llama-3-meets-moe-efficient-upcycling","title":"Llama 3 Meets MoE: Efficient Upcycling","arxiv_id":"2412.09952","date":"2024-12-13","proceeding":null,"authors":["Aditya Vavre","Ethan He","Dennis Liu","Zijie Yan","June Yang","Nima Tajbakhsh","Ashwath Aithal"],"abstract":"Scaling large language models (LLMs) significantly improves performance but comes with prohibitive computational costs. Mixture-of-Experts (MoE) models offer an efficient alternative, increasing capacity without a proportional rise in compute requirements. However, training MoE models from scratch poses challenges like overfitting and routing instability. We present an efficient training recipe leveraging pre-trained dense checkpoints, training an 8-Expert Top-2 MoE model from Llama 3-8B with less than $1\\%$ of typical pre-training compute. Our approach enhances downstream performance on academic benchmarks, achieving a $\\textbf{2%}$ improvement in 0-shot accuracy on MMLU, while reaching a Model FLOPs Utilization (MFU) of $\\textbf{46.8%}$ during training using our framework. We also integrate online upcycling in NeMo for seamless use of pre-trained weights, enabling cost-effective development of high-capacity MoE models.","url_abs":"https://arxiv.org/abs/2412.09952v1","url_pdf":"https://arxiv.org/pdf/2412.09952v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"llama-3-meets-moe-efficient-upcycling","repo_url":"https://github.com/NVIDIA/NeMo","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"mmlu","task_name":"MMLU"},{"task_slug":"mixture-of-experts","task_name":"Mixture-of-Experts"},{"task_slug":"multi-task-language-understanding","task_name":"Multi-task Language Understanding"}],"methods":[{"method_slug":"llama","method_name":"LLaMA"},{"method_slug":"moe","method_name":"MoE"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/multi-task-language-understanding-on-mmlu","task":"Multi-task Language Understanding","dataset":"MML","model":"Llama 3.1 (405B)","rank_in_archive_order":2,"of":44,"metrics":{"Average (%)":"86.6"},"uses_additional_data":true},{"leaderboard":"/sota/multi-task-language-understanding-on-mmlu","task":"Multi-task Language Understanding","dataset":"MML","model":"Llama 3.1 (70B)","rank_in_archive_order":3,"of":44,"metrics":{"Average (%)":"86.0"},"uses_additional_data":true}],"syntology":{"syntology_url":"https://syntology.ai/paper/2412.09952","atlas_url":"https://app.syntology.ai/?focus=2412.09952","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}