act-plus-plus icon indicating copy to clipboard operation
act-plus-plus copied to clipboard
verified publisher icon MarkFzp

When the loss is NaN during training

Open Sslishy opened this issue 11 months ago • 0 comments

I use batch size = 32 ir = 1e-5 who can help me solve this question?thx 2025-01-16_15-54

Sslishy avatar Jan 16 '25 07:01 Sslishy