Research map / Architectures and efficiency / RNNs and normalization
The Stability of Singular Distribution: A Spectral Perspective on the Two-Phase Dynamics of Language Model Pre-training
Hongtao Zhang, WenJie Zhou, Chenxi Jia, Wei Chen, Xueqi Cheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add weight decay (ablation) | AdamW · LLaMA 0.5B pre-training | final validation loss | Improved |
| add weight decay (ablation) | AdamW · LLaMA 0.5B pre-training | SD variation | Improved |
| use Muon instead of Adam | WSD schedule · GPT-2 Small | validation loss | Improved |
| use Muon instead of Adam | WSD schedule · GPT-2 Small | SD variation: 10^-3 -> 10^-4 | Improved |
| decay learning rate by factor 10 at steps 10.2k and 15.3k (ablation) | step-wise schedule · GPT-2 Small | validation loss | Improved |
| start linear decay at step 3650 after stable LR phase (ablation) | WSD · GPT-2 Small | validation loss | Improved |
| use cosine decay (ablation) | LLaMA 0.5B · LLaMA 0.5B pre-training | validation loss | Improved |
Closest papers by meaning
- PolyPythias: Stability and Outliers across Fifty Language Model Pre-Training Runs (ICLR 2025 · similarity 0.89)
- Tracing the Representation Geometry of Language Models from Pretraining to Post-training (NeurIPS 2025 · similarity 0.88)
- Universal One-third Time Scaling in Learning Peaked Distributions (ICML 2026 · similarity 0.88)
- An Empirical Study on Noisy Data and LLM Pretraining Loss Divergence (NeurIPS 2026 · similarity 0.87)
- Prescriptive Scaling Reveals the Evolution of Language Model Capabilities (ICML 2026 · similarity 0.87)
- Through the River: Understanding the Benefit of Schedule-Free Methods for Language Model Training (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
