NorMuon: Making Muon more efficient and scalable — part 3/3

Cleaned-text code points [31551, 33357). Bibliography omitted here; source.md includes it.

For the 124M model, Adam’s parameters uses a learning rate of 3.6 × 10 − 3 3.6\times 10^{-3} with momentum parameters ( β 1 , β 2 ) = ( 0.9 , 0.95 ) (\beta_{1},\beta_{2})=(0.9,0.95) . For Muon and NorMuon, we set the learning rate to 3.6 × 10 − 4 3.6\times 10^{-4} and conduct a grid search over β 1 ∈ { 0.9 , 0.95 } \beta_{1}\in\{0.9,0.95\} , reporting the best result. For NorMuon, β 2 \beta_{2} is set to 0.95.

For the 350M model, Adam employs differentiated learning rates: 0.3 0.3 for the embedding layer and 3 × 10 − 3 3\times 10^{-3} for the output layer, with momentum parameters ( β 1 , β 2 ) = ( 0.8 , 0.95 ) (\beta_{1},\beta_{2})=(0.8,0.95) . For hidden layers, Muon and NorMuon use a learning rate of 7.5 × 10 − 4 7.5\times 10^{-4} , with β 1 \beta_{1} selected from { 0.9 , 0.95 } \{0.9,0.95\} based on validation performance.

## Instructions for reporting errors

We are continuing to improve HTML versions of papers, and your feedback helps enhance accessibility and mobile support. To report errors in the HTML that will help us improve conversion and rendering, choose any of the methods listed below:

Click the "Report Issue" ( ) button, located in the page header.

Tip: You can select the relevant text first, to include it in your report.

Our team has already identified the following issues . We appreciate your time reviewing and reporting rendering errors we may not have found yet. Your efforts will help us improve the HTML versions for all readers, because disability should not be a barrier to accessing research. Thank you for your continued support in championing open access for all.

Have a free development cycle? Help support accessibility at arXiv! Our collaborators at LaTeXML maintain a list of packages that need conversion , and welcome developer contributions .