FoundationVision / LlamaGen

Autoregressive Model Beats Diffusion: 🦙 Llama for Scalable Image Generation

Home Page:https://arxiv.org/abs/2406.06525

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

why using constant lr?

ThisisBillhe opened this issue · comments

Hi authors! Thanks for your excellent work. I just wonder why do you use constant lr rather than other lr_schedulers, which are often preferred.

Looking forward to your reply!

Hi~
The main reason we use constant lr is following DiT.
We didn't try other lr schedulers but I really agree with you that they are often preferred.