mirror of
https://github.com/deepseek-ai/ESFT.git
synced 2026-11-05 00:57:35 +00:00
[GH-ISSUE #19] LoRA configs & scripts #13
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/ESFT#13
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @xfxcwynlc on GitHub (May 12, 2025).
Original GitHub issue: https://github.com/deepseek-ai/ESFT/issues/19
While trying to reproduce the experiment, I’m not able to match the training time reported in the paper. With my current LoRA configuration, the run takes roughly two hours.
Could you please share the LoRA configuration file and the scripts you used for fine-tuning the vanilla model? Thanks a lot :)
@ZihanWang314 commented on GitHub (May 22, 2025):
Our experiments were ran on 2*8 A100 GPUs in the DeepSeek GPU cluster, so it is normal to run longer if you have less than 16 GPUs. We currently do not have the plan to add direct lora training support since all configs are listed in the paper and you can easily implement them with other awesome packages like accelerate:)