You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, I'm curious if I can resume the fine-tuning process from an existing checkpoint?
Right now, I have to start fine-tuning from scratch. Being able to pick up where I left off from an existing checkpoint would be incredibly helpful. Is there a way to achieve this in llama-recipes? I'm using FSDP+PEFT for finetuning. Thanks a lot!
Motivation:
The fine-tuning process stops after specified epochs (e.g., 3 epochs), but later I may find that 3 epochs are not enough and I want to fine-tune for more epochs. Currently, I would have to restart the fine-tuning from scratch. But if I could resume the fine-tuning from the existing checkpoint for epoch 3, then I can save time and resources.
Alternatives
No response
Additional context
No response
The text was updated successfully, but these errors were encountered:
馃殌 The feature, motivation and pitch
Hi, I'm curious if I can resume the fine-tuning process from an existing checkpoint?
Right now, I have to start fine-tuning from scratch. Being able to pick up where I left off from an existing checkpoint would be incredibly helpful. Is there a way to achieve this in llama-recipes? I'm using FSDP+PEFT for finetuning. Thanks a lot!
Motivation:
The fine-tuning process stops after specified epochs (e.g., 3 epochs), but later I may find that 3 epochs are not enough and I want to fine-tune for more epochs. Currently, I would have to restart the fine-tuning from scratch. But if I could resume the fine-tuning from the existing checkpoint for epoch 3, then I can save time and resources.
Alternatives
No response
Additional context
No response
The text was updated successfully, but these errors were encountered: