Hakuli
Copyright 2026 XAEDRO

This product is an independent, clean-room implementation of optimization
concepts described in the following academic publications. No source code
from these projects is included in this distribution:

  - GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection.
    J. Zhao, Z. Zhang, B. Chen, Z. Wang, A. Anandkumar, Y. Tian.
    ICML 2024. arXiv:2403.03507

  - APOLLO: SGD-like Memory, AdamW-level Performance.
    H. Zhu, Z. Zhang, W. Cong, X. Liu, S. Park, V. Chandra, B. Long,
    D. Z. Pan, Z. Wang, J. Lee. MLSys 2025. arXiv:2412.05270

  - Fira: Can We Achieve Full-Rank Training of LLMs Under Low-rank
    Constraint? X. Chen, K. Feng, C. Li, X. Lai, X. Yue, Y. Yuan, G. Wang.
    NeurIPS 2025. arXiv:2410.01623 (norm-growth limiter concept)

  - Q-GaLore: Quantized GaLore with INT4 Projection and Layer-Adaptive
    Low-Rank Gradients. Z. Zhang, A. Jaiswal, L. Yin, S. Liu, J. Zhao,
    Y. Tian, Z. Wang. arXiv:2407.08296, 2024

The citations above are provided for academic attribution of the underlying
research ideas only. Hakuli is not affiliated with, endorsed by, or derived
from the source code of any of these projects. Hakuli was implemented from
published algorithmic descriptions and independently written source code.

"XAEDRO" and "Hakuli" are trademarks of XAEDRO. All other trademarks are
the property of their respective owners.
