Abstract:Low-rank architectures have become increasingly important for efficient large language model (LLM) pre-training, providing substantial reductions in both parameter complexity and memory/computational demands. Despite these advantages, current low-rank methods face three critical shortcomings: (1) compromised model performance, (2) considerable computational overhead, and (3) limited activation memory savings. To address these limitations, we propose Cross-layer Low-Rank residual Network (CR-Net), an innovative parameter-efficient framework inspired by our discovery that inter-layer activation residuals possess low-rank properties. CR-Net implements this insight through a dual-path architecture that efficiently reconstructs layer activations by combining previous-layer outputs with their low-rank differences, thereby maintaining high-rank information with minimal parameters. We further develop a specialized activation recomputation strategy tailored for CR-Net that dramatically reduces memory requirements. Extensive pre-training experiments across model scales from 60M to 7B parameters demonstrate that CR-Net consistently outperforms state-of-the-art low-rank frameworks while requiring fewer computational resources and less memory.
| Comments: | 32 pages. Accepted by ICLR 2026 |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2509.18993 [cs.LG] |
| (or arXiv:2509.18993v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2509.18993 arXiv-issued DOI via DataCite |
Submission history
From: Boao Kong [view email]
[v1]
Tue, 23 Sep 2025 13:43:02 UTC (1,535 KB)
[v2]
Mon, 9 Feb 2026 06:38:03 UTC (1,545 KB)
[v3]
Wed, 13 May 2026 13:30:21 UTC (2,653 KB)