Fix train.py for Adam optimizer betas type error

[mehmoodulhaq570]

Fix Adam Optimizer Betas Type Error

Overview

This update fixes a runtime error encountered during training when initializing the Adam optimizer.

Problem

Training fails with the following error:

ValueError: betas must be either both floats or both Tensors

This happens because the betas parameter is provided as a Python list instead of the required tuple format.

Root Cause

PyTorch’s torch.optim.Adam expects the betas argument in the following form:

betas=(float, float)

However, the existing configuration uses:

betas=[0, 0.99]

When the optimizer is created via:

dnnlib.util.construct_class_by_name(...)

this results in a runtime ValueError.

Fix

The optimizer configuration has been updated to pass betas as a tuple:

betas=(0.0, 0.99)

This fix has been applied to:

• Generator optimizer configuration
• Discriminator optimizer configuration

Impact

• Resolves training crash during optimizer initialization
• Ensures full compatibility with PyTorch Adam
• No change to optimizer behavior or hyperparameter values
• Improves training reliability

Testing

• Confirmed that the Adam optimizer initializes without errors
• Verified that training runs successfully after the fix

[mehmoodulhaq570]

This pull request fixes a runtime error encountered during training when initializing the Adam optimizer.

The issue was caused by passing the betas parameter as a Python list ([0, 0.99]), whereas PyTorch’s torch.optim.Adam requires betas to be provided as a tuple of floats.

[BStandage]

I encountered the same Adam optimizer betas type error during training (CUDA 12.4, PyTorch 2.6.0). Converting betas to a tuple of floats resolves it.

Suggestion: keeping the change stylistically consistent with the surrounding code (inline EasyDict without additional comments) may help keep the diff minimal and aligned with existing formatting.

Example:

c.G_opt_kwargs = dnnlib.EasyDict(class_name='torch.optim.Adam', betas=(0.0,0.99), eps=1e-8)
c.D_opt_kwargs = dnnlib.EasyDict(class_name='torch.optim.Adam', betas=(0.0,0.99), eps=1e-8)