Free tool · browser-only calculation
Learning rate scheduler visualizer with PyTorch code
Visualise and customise learning rate schedules with warmup, cosine, one-cycle, step and copyable PyTorch code. With guided examples and shareable.
How the calculation works
Formula as plain text
\eta_t=\eta_{min}+\frac{1}{2}(\eta_{max}-\eta_{min})\left(1+\cos\left(\frac{t\pi}{T}\right)\right)A scheduler changes the learning rate during training. Constant values simplify comparison, step and multistep reduce it at intervals, while exponential and polynomial schedules vary it gradually. These choices affect optimisation speed and stability.
Cosine annealing follows a cosine curve, warmup increases the rate initially and one-cycle combines rising and falling phases. The chart exposes slope changes and effective values per step or epoch.
Equivalent PyTorch code makes the configuration reusable. Examples can be edited and shared through the URL; no training is executed and model quality is not evaluated. Use the visualiser to compare schedules before testing them on a validation set.