Research map / Reinforcement learning / RL from feedback
Temperature as a Meta-Policy: Adaptive Temperature in LLM Reinforcement Learning
Haoran Dang, Cuiling Lan, Hai Wan, Xibin Zhao, Yan Lu
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 |
|---|---|---|---|
| learn a temperature meta-policy (TAMPO) instead of fixed temperature in GRPO | GRPO on DS-Qwen-1.5B · 5 math benchmarks avg | Pass@1: 42.6 (best fixed T=1.5) -> 44.5 | Improved |
| learn a temperature meta-policy instead of fixed temperature in GRPO | GRPO on DS-Qwen-1.5B · 5 math benchmarks avg | Pass@8: 62.1 -> 63.8 | Improved |
| learn a temperature meta-policy instead of linearly increasing temperature | GRPO · OlympiadBench | Pass@1: 41.0 -> 39.6 | Hurt |
| raise EMA coefficient alpha from 0.01 to 0.05 (ablation) | TAMPO · 5 math benchmarks avg | Pass@1: 41.6 -> 44.5 | Improved |
| raise EMA coefficient alpha from 0.05 to 0.10 (ablation) | TAMPO · 5 math benchmarks avg | Pass@1: 44.5 -> 43.6 | Mixed |
| use greedy selection of temperature from meta-policy instead of top-p 0.7 (ablation) | TAMPO · 5 math benchmarks avg | Pass@1: 44.5 -> 40.9 | Hurt |
| use top-p 0.9 instead of 0.7 for meta-policy sampling (ablation) | TAMPO · 5 math benchmarks avg | Pass@1: 44.5 -> 43.0 | Hurt |
| apply TAMPO instead of GRPO with a different base model and task | Qwen2.5-3B-Instruct · ECQA | Pass@1: 75.07% -> 76.12% | Improved |
Closest papers by meaning
- TGRL: Temperature-Grouped Reinforcement Learning for Efficient Exploration in LLMs (NeurIPS 2026 · similarity 0.92)
- TempSamp-R1: Effective Temporal Sampling with Reinforcement Fine-Tuning for Video LLMs (NeurIPS 2025 · similarity 0.88)
- LLM-Explorer: A Plug-in Reinforcement Learning Policy Exploration Enhancement Driven by Large Language Models (NeurIPS 2025 · similarity 0.88)
- Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs (NeurIPS 2025 · similarity 0.88)
- BOTS: A Unified Framework for Bayesian Online Task Selection in LLM Reinforcement Finetuning (ICLR 2026 · similarity 0.87)
- EVOLvE: Evaluating and Optimizing LLMs For In-Context Exploration (ICML 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.
