Research map / Large language models / Agents, code and math
Trinity: An Evolved LLM Coordinator
Jinglue Xu, Qi Sun, Peter Schwendeman, Stefan Nielsen, Edoardo Cetin, Yujin Tang
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 singular value fine-tuning of the coordinator SLM (ablation) | Trinity · In-distribution average | average score: 67.85 -> 70.44 | Improved |
| add thinker-role selection (ablation) | Trinity · In-distribution average | average score: 68.69 -> 70.44 | Improved |
| add tri-role selection (ablation) | Trinity · In-distribution average | average score: 67.02 -> 70.44 | Improved |
| add tri-role selection (ablation) | Trinity · MATH500 | accuracy: 82.00 -> 88.00 | Improved |
| use penultimate token instead of last token for hidden state (ablation) | Trinity · In-distribution average | average score: 64.66 -> 70.44 | Improved |
| use penultimate token instead of last token (ablation) | Trinity · LiveCodeBench | pass@1 (%): 50.85 -> 61.46 | Improved |
| add agent selection to role selection (ablation) | Trinity · In-distribution average | average score: 65.95 -> 70.44 | Improved |
| use linear head instead of low-rank head (ablation) | Trinity · MATH500 | accuracy: 0.770 -> 0.880 | Improved |
| use sparse head instead of linear head (ablation) | Trinity · MMLU | accuracy: 0.916 -> 0.917 | No clear effect |
| use sep-CMA-ES instead of REINFORCE (ablation) | Trinity · LiveCodeBench | pass@1: 0.253 -> 0.615 | Improved |
| use sep-CMA-ES instead of SFT (ablation) | Trinity · MATH500 | accuracy: 0.786 -> 0.880 | Improved |
| use sep-CMA-ES instead of random search (ablation) | Trinity · RLPR | accuracy: 0.345 -> 0.401 | Improved |
Closest papers by meaning
- MergeBench: A Benchmark for Merging Domain-Specialized LLMs (NeurIPS 2025 · similarity 0.89)
- LLM Augmented LLMs: Expanding Capabilities through Composition (ICLR 2024 · similarity 0.89)
- Unveiling the Secret Recipe: A Guide For Supervised Fine-Tuning Small LLMs (ICLR 2025 · similarity 0.88)
- Language Modeling by Language Models (NeurIPS 2025 · similarity 0.87)
- Automated Creation of Reusable and Diverse Toolsets for Enhancing LLM Reasoning (AAAI 2025 · similarity 0.87)
- PACE: Two-Timescale Self-Evolution for Small Language Model Agents (NeurIPS 2026 · 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.
