Skip to content

Validate LSTM/GRU full-sequence output regardless of the returnSequence flag #949

Description

@shiyi9801

Some underlying backends allocate the full-sequence output tensor internally regardless of the returnSequence flag but do not validate its size, which might cause an out-of-bounds write. We should always validate the full-sequence output descriptor [steps, numDirections, batchSize, hiddenSize] so that oversized shapes are rejected before reaching the backend.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Agenda+security-trackerGroup bringing to attention of security, or tracked by the security Group but not needing response.

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions