Skip to content

Conversation

@ben-z
Copy link
Contributor

@ben-z ben-z commented Jul 14, 2025

What this does

Similar to #1198 , pi0fast's loss is also of type Tensor, which is not accepted by wandb. This PR converts it to a scalar using .item()

How it was tested

Manually ran training and verified that loss is now visible in the wandb dashboard for pi0fast.

How to checkout & try? (for the reviewer)

Train pi0fast (instructions) and observe the loss log in wandb.

@ben-z ben-z force-pushed the benz/pi0fast-loss branch from b1fbe77 to 4f0eaec Compare July 14, 2025 00:43
Copy link
Collaborator

@AdilZouitine AdilZouitine left a comment

Choose a reason for hiding this comment

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

LGTM

@CarolinePascal
Copy link
Collaborator

Hi @ben-z !

Small update : we will put this PR on hold for a while, as we first need to bring some sense and coherence in the loss format across policies (forward() never has the same return type 🫠).

Best,

Caroline.

@AdilZouitine AdilZouitine added bug Something isn’t working correctly policies Items related to robot policies visualization Issues about visual output, graphs, or data visualization labels Aug 12, 2025
@imstevenpmwork
Copy link
Collaborator

Closing this as PI0Fast has been deprecated. Thanks !

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

Labels

bug Something isn’t working correctly policies Items related to robot policies visualization Issues about visual output, graphs, or data visualization

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants