Research map / Architectures and efficiency / Attention and transformers
The Expressive Power of Low Precision Softmax Transformers with (Summarized) Chain-of-Thought
Moritz Brösamle, Stephan Eckstein
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 |
|---|---|---|---|
| increase CoT model depth from 6 to 8 (ablation) | CoT transformer on Sudoku · Sudoku-extreme, trained on CoT length up to 2^14 | accuracy at long context | Improved |
| raise activation precision from bfloat16 to fp32 (ablation) | CoT transformer depth 6 · Sudoku-extreme, trained on CoT length up to 2^14 | accuracy at long context | No clear effect |
| use summarized CoT instead of plain CoT at the same model size | depth-6, dim-512 transformer · Sudoku-extreme long contexts | accuracy | Improved |
| increase SCoT depth from 4 to 5 (ablation) | SCoT transformer dim 512 · Sudoku-extreme | accuracy (%): 16.3 -> 97.2 | Improved |
| increase SCoT depth beyond 6 (ablation) | SCoT transformer dim 512 · Sudoku-extreme | accuracy (%): 99.5 -> 99.7 -> 99.6 | No clear effect |
| longer training for N=2^14 CoT run (ablation) | CoT depth 6 · Sudoku-extreme | accuracy at long context | Mixed |
Closest papers by meaning
- Chain of Thought Empowers Transformers to Solve Inherently Serial Problems (ICLR 2024 · similarity 0.89)
- Pause Tokens Strictly Increase the Expressivity of Constant-Depth Transformers (NeurIPS 2025 · similarity 0.89)
- Strassen Attention, Split VC Dimension and Compositionality in Transformers (NeurIPS 2025 · similarity 0.89)
- Learning Linear Attention in Polynomial Time (NeurIPS 2025 · similarity 0.88)
- The Expressive Power of Transformers with Chain of Thought (ICLR 2024 · similarity 0.88)
- Lower Bounds for Chain-of-Thought Reasoning in Hard-Attention Transformers (ICML 2025 · similarity 0.88)
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.
