Research map / Image generation / Autoregressive and latent generation
Long Horizon Temperature Scaling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use joint LHTS rather than myopic temperature scaling | GPT-2 language models · Text generation, diversity-likelihood trade-off | sample diversity and log-likelihood | Helped |
| Use LHTS instead of diffusion noise-variance pseudo-temperature scaling | DDPM · CIFAR-10 | FID: 3.94 -> 3.66 | Helped |
| Use LHTS instead of diffusion noise-variance pseudo-temperature scaling | DDPM · CIFAR-10 | log-likelihood: -3.09 -> -3.07 | Helped |
| Tune long-horizon temperature alongside myopic temperature | Text8 character model · Character sample likelihood and diversity | log-likelihood: -1.05 -> -0.97 | Helped |
| Apply LHTS at myopic T=0.0 | GPT-2 large · Analogy multiple choice, duplicate questions | accuracy: 0.290 -> 0.317 | Helped |
| Apply LHTS at myopic T=0.0 | GPT-2 large · Analogy multiple choice, distinct questions | accuracy: 0.250 -> 0.232 | Hurt |
| Apply LHTS to GPT-2 large | GPT-2 large · OpenWebText | MAUVE: 0.76 -> 0.41 | Hurt |
| Add an auxiliary KL loss to LHTS finetuning (ablation) | LHTS language model · Language model experiments | performance | No clear effect |
| Use LHTS rather than myopic temperature scaling | GPT-2 large · HELM XSUM test | xsum_test_rouge2: 0.016 -> 0.02 | Helped |
| Use LHTS rather than myopic temperature scaling | GPT-2 large · HELM BoolQ | boolq_exact_match: 0.583 -> 0.603 | Helped |
| Use LHTS rather than myopic temperature scaling | GPT-2 large · HELM NaturalQA open book | naturalqa_open_f1_score: 0.257 -> 0.248 | Hurt |
| Scale samples using temperature (myopic or LHTS) rather than no scaling | GPT-2 large · OpenWebText | MAUVE: 0.76 -> 0.57; 0.76 -> 0.41 | Hurt |
Closest papers by meaning
- Optimizing Temperature for Language Models with Multi-Sample Inference (ICML 2025 · similarity 0.87)
- Scaling Inference Time Compute for Diffusion Models (CVPR 2025 · similarity 0.86)
- On the Limitations of Temperature Scaling for Distributions with Overlaps (ICLR 2024 · similarity 0.86)
- Inference-Time Scaling for Flow Models via Stochastic Generation and Rollover Budget Forcing (NeurIPS 2025 · similarity 0.85)
- Uncertainty Quantification for Large Language Diffusion Models (NeurIPS 2026 · similarity 0.84)
- Scale Where It Matters: Training-Free Localized Scaling for Diffusion Models (NeurIPS 2026 · similarity 0.84)
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.