Research map / Architectures and efficiency / Pruning and efficient tuning
Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer
Greg Yang, Edward J Hu, Igor Babuschkin, Szymon Sidor, Xiaodong Liu, David Farhi, Nick Ryder, Jakub Pachocki, Weizhu Chen, Jianfeng Gao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| transfer HPs from 0.25x proxy via muP instead of tuning the 1x model directly (equal compute) | IWSLT14 Transformer · IWSLT14 De-En | median val BLEU: 35.00 -> 35.33 | Improved |
| use muP parametrization instead of SP when transferring HPs (ablation) | IWSLT14 Transformer · IWSLT14 De-En | BLEU: training diverged -> 35.33 | Improved |
| muTransfer vs tuning 1x with same number of samples | IWSLT14 Transformer · IWSLT14 De-En | BLEU | Mixed |
| muTransfer from 15M proxy | WMT14 Transformer 211M · WMT14 En-De | median val BLEU: 25.69 -> 26.34 | Improved |
| muTransfer from BERT-prototype | BERT-large · BERT pretraining | test loss: 1.731 -> 1.683 | Improved |
| muTransfer from BERT-prototype | BERT-base · BERT pretraining | test loss: 1.995 -> 1.970 | Improved |
| muTransfer HPs on GPT-3 6.7B | GPT-3 6.7B · HellaSwag Zero-Shot | accuracy: 66.7 -> 72.0 | Improved |
| muTransfer HPs on GPT-3 6.7B | GPT-3 6.7B · LAMBADA | accuracy: Zero-shot 70.8 -> 73.5; Few-shot 77.1 -> 74.7 | Mixed |
| transfer best init std across depth | 2-layer Transformer muP · Wikitext-2 | loss optimum | Hurt |
Closest papers by meaning
- On the Provable Separation of Scales in Maximal Update Parameterization (ICML 2025 · similarity 0.90)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · similarity 0.89)
- Sparse maximal update parameterization: A holistic approach to sparse training dynamics (NeurIPS 2024 · similarity 0.87)
- Sample complexity of data-driven tuning of model hyperparameters in neural networks with structured parameter-dependent dual function (NeurIPS 2025 · similarity 0.87)
- Scaling Exponents Across Parameterizations and Optimizers (ICML 2024 · similarity 0.87)
- u-$\mu$P: The Unit-Scaled Maximal Update Parametrization (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
