Skip to content

Conversation

@craymichael
Copy link
Contributor

Summary: As title. It is possible for attr to be computed as an estimated amount over multiple samples of the response, so the estimate has variance. This adds an attribute to store this variance in the results, if we have it.

Differential Revision: D84970183

@meta-codesync
Copy link
Contributor

meta-codesync bot commented Oct 25, 2025

@craymichael has exported this pull request. If you are a Meta employee, you can view the originating Diff in D84970183.

Summary:

Python 3.9 is EOL. Bump minimum version of Python to 3.10.

Differential Revision: D85569426
…eta-pytorch#1657)

Summary:

Refactor LLMAttributionResult into an abstract base object that is generic. Create LLMAttributionResult as a concrete child with aliases for captum.attr API supporting legacy use. Changes support the refactor and enable more generalized use beyond logprob-based attribution.

Differential Revision: D84721127
…-pytorch#1658)

Summary:

Update LLM attr definition to accommodate other typing considerations. Clean up some variable names as well.

Differential Revision: D84721071
…ance to LLM attribution results (meta-pytorch#1659)

Summary:

As title. It is possible for attr to be computed as an estimated amount over multiple samples of the response, so the estimate has variance. This adds an attribute to store this variance in the results, if we have it.

Differential Revision: D84970183
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant