[2203.15556] Training Compute-Optimal Large Language Models

Abstract:We investigate the optimal model size and number of tokens for training a transformer language model under a given compute budget. We find that current large language models are significantly undertrained, a consequence of the recent focus on scaling language models whilst keeping the amount of training data constant. By training over 400 language models ranging from 70 million to over 16 billion parameters on 5 to 500 billion tokens, we find that for compute-optimal training, the model size and the number of training tokens should be scaled equally: for every doubling of model size the number of training tokens should also be doubled. We test this hypothesis by training a predicted compute-optimal model, Chinchilla, that uses the same compute budget as Gopher but with 70B parameters and 4$\times$ more more data. Chinchilla uniformly and significantly outperforms Gopher (280B), GPT-3 (175B), Jurassic-1 (178B), and Megatron-Turing NLG (530B) on a large range of downstream evaluation tasks. This also means that Chinchilla uses substantially less compute for fine-tuning and inference, greatly facilitating downstream usage. As a highlight, Chinchilla reaches a state-of-the-art average accuracy of 67.5% on the MMLU benchmark, greater than a 7% improvement over Gopher.

[2203.15556] Training Compute-Optimal Large Language Models Skip to main content arXiv is now an independent nonprofit! Learn more × Search arXiv Press Enter to search · Advanced search --> Computer Science > Computation and Language arXiv:2203.15556 (cs) [Submitted on 29 Mar 2022] Title: Training Compute-Optimal Large Language Models Authors: Jordan Hoffmann , Sebastian Borgeaud , Arthur Mensch , Elena Buchatskaya , Trevor Cai , Eliza Rutherford , Diego de Las Casas , Lisa Anne Hendricks , Johannes Welbl , Aidan Clark , Tom Hennigan , Eric Noland , Katie Millican , George van den

related reading

• Chinchillaarxiv.org
• [2510.13786] The Art of Scaling Reinforcement Learning Compute for LLMsarxiv.org
• [2005.14165] Language Models are Few-Shot Learnersarxiv.org
• [2309.10668] Language Modeling Is Compressionarxiv.org
• [2106.09685] LoRA: Low-Rank Adaptation of Large Language Modelsarxiv.org
• [2509.14786] Pre-training under infinite computearxiv.org
• [2203.02155] Training language models to follow instructions with human feedbackarxiv.org
• Transformers Inference Optimization Toolset | AstraBlogastralord.github.io
• IsoCompute Playbook: Optimally Scaling Sampling Compute for RL Training of LLMscompute-optimal-rl-llm-scaling.github.io
• [2605.12715] Scaling Laws for Mixture Pretraining Under Data Constraintsarxiv.org
• New Scaling Laws for Large Language Models — LessWronglesswrong.com
• [2605.06546] Efficient Pre-Training with Token Superpositionarxiv.org