Rahul Chalamala Research

Dragonfly: Multi-Resolution Zoom Supercharges Large Visual-Language Model

Kezhen Chen, Rahul Thapa, Rahul Chalamala, Ben Athiwaratkun, Shuaiwen Leon Song, James Zou
Preprint. Under review.
Dragonfly is a new large multimodal model architecture that enhances fine-grained visual understanding and reasoning about image regions using multi-resolution visual encoding and zoom-in patch selection. It achieves state-of-the-art results on multiple benchmarks, including biomedical tasks, demonstrating its effectiveness and versatility.

    LeanDojo: Theorem Proving with Retrieval-Augmented Language Models

    Kaiyu Yang, Aidan Swope, Alex Gu, Rahul Chalamala, Peiyang Song, Shixing Yu, Saad Godil, Ryan Prenger, Anima Anandkumar
    Neural Information Processing Systems (NeurIPS) - Oral Presentation
    LeanDojo is an open-source Lean playground that provides tools, data, models, and benchmarks for LLM-based theorem proving. It features ReProver, a retrieval-augmented prover, and a comprehensive benchmark of 98,734 theorems to facilitate research and development.

      Spectrum Safety: Compatibility of NTS-3 Signals with GNSS Signals.

      Rahul Chalamala, Joanna Hinks
      Proceedings of the ION 2022 Joint Navigation Conference - Oral Presentation
      We developed a Python-based framework to assess the potential interference of NTS-3 signals with GNSS signals, using Spectral Separation Coefficients and ITU-R guidelines. Our preliminary findings show minimal interference, and we have proposed several strategies to mitigate any potential issues.