Skip to content

šŸ” Add guidance on choosing max_length value and include visualizatiā€¦ #3630

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. Weā€™ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

šŸ” Add guidance on choosing max_length value and include visualizatiā€¦ #3630

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b8a0080
šŸ” Add guidance on choosing `max_length` value and include visualizatiā€¦
qgallouedec Jun 22, 2025
f016018
šŸ” Update iframe width for dataset length profiler visualization
qgallouedec Jun 22, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions docs/source/reducing_memory_usage.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,14 @@ training_args = SFTConfig(..., max_length=...)
</hfoption>
</hfoptions>

### How to choose the `max_length` value?

If `max_length` is too small, a significant portion of your tokens will be discarded and won't contribute to training. If it's too large, memory usage can spike, potentially leading to OOM (Out-Of-Memory) errors. Without packing, a large `max_length` may also result in inefficient training, as many tokens will be padding.

To help you choose an appropriate value, we provide a utility to visualize the sequence length distribution in your dataset.

<iframe src="https://trl-lib-dataset-length-profiler.hf.space" frameborder="0" width="100%" height="1000"></iframe>

## Packing

<Tip>
Expand Down