LLaMA-Factory
LLaMA-Factory copied to clipboard
support kimi-vl in zero3
Reminder
- [x] I have read the above rules and searched the existing issues.
Description
I have tried to run the full params training with the use of zero2(8*A100), but met overflow. Is there any plan for the support of kimi-vl zero3 in the future.
Pull Request
No response