LongContextMamba. MambaExtend is a training-free approach designed to enhance the long-context understanding of Mamba-based models. It leverages calibrated scaling factors within the discretization module of the Mamba model. This work has been accepted to ICLR 2025.

github.com/ArminAzizi98/LongContextMamba

Problem

Audience

Updates

No founder updates yet. Check back soon.