Research map / Vision-language and multimodal / CLIP and zero-shot
HTLM: Hyper-Text Pre-Training and Prompting of Language Models
Armen Aghajanyan, Dmytro Okhonko, Mike Lewis, Mandar Joshi, Hu Xu, Gargi Ghosh, Luke Zettlemoyer
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 |
|---|---|---|---|
| Use manually engineered HTML prompts with size hints rather than PEGASUS zero-shot prompting | HTLM · Gigaword | ROUGE-L: 20.20 -> 28.60 | Improved |
| Use manually engineered HTML prompts with size hints rather than PEGASUS zero-shot prompting | HTLM · CNN/DM | ROUGE-1: 32.90 -> 38.51 | Improved |
| Use manually engineered HTML prompts with size hints rather than PEGASUS zero-shot prompting | HTLM · Reddit TIFU | ROUGE-1: 14.66 -> 15.81 | Improved |
| Use manually engineered HTML prompts with size hints rather than PEGASUS zero-shot prompting | HTLM · XSum | ROUGE-1: 19.27 -> 22.34 | Improved |
| Add size hints to auto-prompting (ablation) | HTLM auto-prompting · Gigaword | ROUGE-1: 27.56 -> 28.73 | Improved |
| Add size hints to auto-prompting (ablation) | HTLM auto-prompting · CNN/DM | ROUGE-1: 33.40 -> 34.65 | Improved |
| Add size hints to auto-prompting (ablation) | HTLM auto-prompting · Reddit TIFU | ROUGE-1: 6.71 -> 8.15 | Improved |
| Add size hints to auto-prompting (ablation) | HTLM auto-prompting · XSum | ROUGE-1: 15.15 -> 17.14 | Improved |
| Use HTML-based prompting instead of GPT-3 Med prompting | HTLM-Manual · RTE | accuracy: 49.8 -> 51.2 | Improved |
| Use HTML-based prompting instead of GPT-3 Med prompting | HTLM-Manual · BoolQ | accuracy: 60.3 -> 55.3 | Hurt |
| Add R3F during HTLM fine-tuning (ablation) | HTLM · RTE | accuracy: 87.1 -> 89.1 | Improved |
| Add prompt-based fine-tuning after R3F (ablation) | HTLM-R3F · RTE | accuracy: 89.1 -> 89.4 | Improved |
Closest papers by meaning
- Web2Code: A Large-scale Webpage-to-Code Dataset and Evaluation Framework for Multimodal LLMs (NeurIPS 2024 · similarity 0.86)
- Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights (NeurIPS 2025 · similarity 0.86)
- Beyond Prompt Engineering: A Reinforced Token-Level Input Refinement for Large Language Models (AAAI 2025 · similarity 0.85)
- Generative Adapter: Contextualizing Language Models in Parameters with A Single Forward Pass (ICLR 2025 · similarity 0.85)
- Enhancing Multilingual LLM Pretraining with Model-Based Data Selection (NeurIPS 2025 · similarity 0.85)
- DataMan: Data Manager for Pre-training Large Language Models (ICLR 2025 · 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.
