Fix auto padding free logic to respect user passed False by mmathew23 · Pull Request #4128 · unslothai/unsloth

mmathew23 · 2026-02-27T21:21:18Z

Qwen3-14B notebook currently OOM's on a T4 due to increase VRAM when padding free is turned on. SFTConfig currently defaults padding_free=False so the current logic can't differentiate between when it should autopad or the user specifically requested to turn off.

This PR patches SFTConfig to default padding_free to None. If it's None padding free will be auto-enabled (the default). If it's True it's enabled, and if it's False it's turned off.

Notebook before fix with padding_free=False
https://colab.research.google.com/drive/1u51CbHLntgBLUrWe4B4lZFRNtPGi1faG?usp=sharing

Working notebook after fix with padding_free=False
https://colab.research.google.com/drive/1GYBXNlm9yP8zP0XAT6LD0kmmL5NficK-?usp=sharing

gemini-code-assist · 2026-02-27T21:21:34Z

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request addresses an Out-Of-Memory issue encountered when padding_free was explicitly set to False in the SFTConfig. The core problem stemmed from the inability to distinguish between a user intentionally disabling padding-free and the system's default behavior. By changing the default padding_free value to None, the system can now correctly interpret user intent, allowing auto-padding to be enabled only when not explicitly disabled, thereby preventing unexpected memory consumption.

Highlights

Padding-free logic fix: Resolved an Out-Of-Memory (OOM) issue occurring when padding_free was explicitly set to False in SFTConfig, which previously led to unintended auto-padding.
SFTConfig default change: Modified the default value of padding_free in SFTConfig from False to None to clearly differentiate between a user explicitly disabling it and the system's default behavior.
Auto-padding detection: Updated the internal logic to correctly interpret padding_free=None as a signal to auto-enable padding-free, while respecting True for explicit enablement and False for explicit disablement.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

unsloth/models/rl.py
- Changed the default value of padding_free to None in the _patch_trl_rl_trainers function.
unsloth/trainer.py
- Updated the _should_auto_padding_free function to check if padding_free is None for auto-enabling.
- Added a comment in new_init to clarify the resolution logic for padding_free.

Activity

No specific activity (comments, reviews, etc.) has been recorded for this pull request yet.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request correctly addresses an issue where explicitly setting padding_free=False was not respected, by changing the default value to None and updating the detection logic. The changes look good. I've added one suggestion to improve the robustness of the new check in _should_auto_padding_free.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e9981c3b5b

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-02-27T21:25:02Z

    ):
        return False
-    return not getattr(config, "padding_free", False)
+    return getattr(config, "padding_free", False) is None


Treat missing padding_free as unset for auto mode

The new return getattr(config, "padding_free", False) is None check disables auto padding-free for configs that do not define padding_free at all, because the fallback now becomes False instead of “unset”. This regresses the backwards-compatibility path where callers pass plain transformers.TrainingArguments (handled in _backwards_compatible_trainer), since those configs typically lack this field and will no longer get the previous auto-enable behavior unless users manually inject padding_free.

Useful? React with 👍 / 👎.

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

danielhanchen

Tested and verified. The fix correctly distinguishes between "user didn't set padding_free" (None, auto-enable) and "user explicitly set padding_free=False" (respect it).

Unit tests (7/7 passed): All 4 padding_free states (None, True, False, missing attribute) plus edge cases.

GPU benchmarks (Llama-3.2-1B-Instruct, 61 steps):

Scenario	Peak Mem	train_loss	Stdout message
Baseline (main)	1.53 GB	1.3789	"Padding-free auto-enabled"
PR: default (None)	1.53 GB	1.3789	"Padding-free auto-enabled"
PR: explicit False	1.62 GB	1.3748	No padding-free message
PR: explicit True	1.53 GB	1.3789	"Padding-free enabled"

Baseline vs PR default: losses/grad-norms identical -- zero regression
Baseline vs PR True: losses/grad-norms identical
PR explicit False: correctly disables padding-free (slightly higher memory, slightly different losses from different batching)

No additional changes needed.

) * Fix auto padding free logic to respect user passed * Update unsloth/trainer.py Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --------- Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

Fix auto padding free logic to respect user passed

e9981c3

mmathew23 requested review from Datta0, danielhanchen and pluesclues as code owners February 27, 2026 21:21

gemini-code-assist Bot reviewed Feb 27, 2026

View reviewed changes

Comment thread unsloth/trainer.py Outdated

chatgpt-codex-connector Bot reviewed Feb 27, 2026

View reviewed changes

Update unsloth/trainer.py

316c9d6

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

mmathew23 mentioned this pull request Feb 28, 2026

add padding_free=False to Qwen3 14B notebooks but not Alpaca unslothai/notebooks#202

Merged

danielhanchen approved these changes Mar 1, 2026

View reviewed changes

danielhanchen merged commit 9b56d63 into unslothai:main Mar 2, 2026
1 check passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix auto padding free logic to respect user passed False#4128

Fix auto padding free logic to respect user passed False#4128
danielhanchen merged 2 commits into
unslothai:mainfrom
mmathew23:fix/padding-free-false

mmathew23 commented Feb 27, 2026

Uh oh!

gemini-code-assist Bot commented Feb 27, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Feb 27, 2026

Uh oh!

danielhanchen left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

mmathew23 commented Feb 27, 2026

Uh oh!

gemini-code-assist Bot commented Feb 27, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

danielhanchen left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants