Research map / Architectures and efficiency / Attention and transformers
Tracr: Compiled Transformers as a Laboratory for Interpretability
David Lindner, Janos Kramar, Sebastian Farquhar, Matthew Rahtz, Thomas McGrath, Vladimir Mikulik
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 |
|---|---|---|---|
| compress residual stream from D=14 to d=6 with learned projection W | Tracr frac_prevs model · frac_prevs | loss: 14 -> 6 | No clear effect |
| reduce embedding dimension d below 6 (ablation) | compressed frac_prevs model · frac_prevs | loss | Hurt |
| learned compression matrix instead of PCA with 8 components (ablation) | compressed frac_prevs model · frac_prevs | retained task-relevant features | Improved |
| compress sort_unique with gradient descent | Tracr sort_unique model · sort_unique | layer-output cosine similarity: 0.8 | Hurt |
Closest papers by meaning
- Beyond Components: Singular Vector-Based Interpretability of Transformer Circuits (NeurIPS 2025 · similarity 0.86)
- The Expressive Power of Transformers with Chain of Thought (ICLR 2024 · similarity 0.85)
- Algorithmic Capabilities of Random Transformers (NeurIPS 2024 · similarity 0.85)
- Compact Proofs of Model Performance via Mechanistic Interpretability (NeurIPS 2024 · similarity 0.85)
- Chain of Thought Empowers Transformers to Solve Inherently Serial Problems (ICLR 2024 · similarity 0.85)
- Transcoders find interpretable LLM feature circuits (NeurIPS 2024 · similarity 0.85)
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.
