Skip to content

Dropout scaling can overflow for p values close to 1 #38

Description

@github-actions

File: leanpass/nn.py

Dropout.forward scales the mask by 1/(1-p). When p is very close to 1 (e.g., 0.9999) the scaling factor becomes huge, leading to overflow or exploding activations.

Fix: enforce a stricter upper bound (e.g., p <= 0.9) or use a numerically stable inverted‑dropout implementation that caps the scaling factor.

Label: bug

Filed automatically by ai-issue-scan.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions