does support train lora in 4090 with HunYuan model?

[syyxsxx]

System Info / 系統信息

diffusers: from source

Information / 问题信息

• The official example scripts / 官方的示例脚本
• My own modified scripts / 我自己修改的脚本和任务

Reproduction / 复现过程

i have added --precompute_conditions and --gradient_checkpointing with training lora by hunyuan model on 4090, but it still oom
does support train lora on 4090 with HunYuan model?

Expected behavior / 期待表现

问题得到解决

[a-r-r-o-w]

Unfortunately, it still requires a 48 GB GPU unless one trains with fp8. However, with the new Trainer implementation we are building, fp8 support has not yet been added. We will try and prioritize this as soon as possible!

[syyxsxx]

Thank you for your reply. I'm looking forward to the FP8 training feature.

[a-r-r-o-w]

Hi. We just merged #184, which should allow training under 24 GB upto certain resolution/frame limits. LMK if you face any problems 🤗