Low-rank pre-training methods converge to geometrically and spectrally distinct basins and show diverging activations compared to full-rank training at 60M-350M scales.
Grass: Com- pute efficient low-memory llm training with structured sparse gradients.arXiv preprint arXiv:2406.17660, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Beyond Perplexity: A Geometric and Spectral Study of Low-Rank Pre-Training
Low-rank pre-training methods converge to geometrically and spectrally distinct basins and show diverging activations compared to full-rank training at 60M-350M scales.