Skip to content

[rl/unified] Update default model-ckpt-path in infer.py to the one from README#2405

Merged
daniellepintz merged 1 commit intomainfrom
dp/infer
Feb 20, 2026
Merged

[rl/unified] Update default model-ckpt-path in infer.py to the one from README#2405
daniellepintz merged 1 commit intomainfrom
dp/infer

Conversation

@daniellepintz
Copy link
Contributor

  • Update default model-ckpt-path in infer.py to the one from README (torchtitan/experiments/rl/example_checkpoint/Qwen3-0.6B/)
  • Update README to remove passing --model-ckpt-path to make life a little easier for users.

@daniellepintz daniellepintz requested a review from wwwjn February 20, 2026 12:13
@meta-cla meta-cla bot added the CLA Signed This label is managed by the Meta Open Source bot. label Feb 20, 2026
Copy link
Contributor

@Lucaskabela Lucaskabela left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Definitely makes things easier when setting up :)

@daniellepintz daniellepintz merged commit b36efaf into main Feb 20, 2026
16 of 22 checks passed
4. Run inference:
```bash
python torchtitan/experiments/rl/unified/infer.py --model-ckpt-path <path_to_model_checkpoint>
python torchtitan/experiments/rl/unified/infer.py
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I explicit put the model-ckpt-path here to make the user define which model checkpoint he wants to use. Otherwise infer.py are too tighted to 0.6B model (although the model registry is specific to 0.6B model, which will be alleviate soon int the PR stack), and user might not realize which checkpoint path they are using.

I would prefer leave --model-ckpt-path here. Other changes looks good

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, added back in #2424

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants