Papers › HTLM: Hyper-Text Pre-Training and Prompting of Language Models

HTLM: Hyper-Text Pre-Training and Prompting of Language Models

14 Jul 2021ICLR 2022 4arXiv:2107.06955archive 2025-07-28

Armen Aghajanyan, Dmytro Okhonko, Mike Lewis, Mandar Joshi, Hu Xu, Gargi Ghosh, Luke Zettlemoyer

We introduce HTLM, a hyper-text language model trained on a large-scale web crawl. Modeling hyper-text has a number of advantages: (1) it is easily gathered at scale, (2) it provides rich document-level and end-task-adjacent supervision (e.g. class and id attributes often encode document category information), and (3) it allows for new structured prompting that follows the established semantics of HTML (e.g. to do zero-shot summarization by infilling title tags for a webpage that contains the input text). We show that pretraining with a BART-style denoising loss directly on simplified HTML provides highly effective transfer for a wide range of end tasks and supervision levels. HTLM matches or exceeds the performance of comparably sized text-only LMs for zero-shot prompting and fine-tuning for classification benchmarks, while also setting new state-of-the-art performance levels for zero-shot summarization. We also find that hyper-text prompts provide more value to HTLM, in terms of data efficiency, than plain text prompts do for existing LMs, and that HTLM is highly effective at auto-prompting itself, by simply generating the most likely hyper-text formatting for any available training data. We will release all code and models to support future HTLM research.

PaperPDFConference PDF

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Data-to-Text GenerationDenoisingLanguage ModellingTable-to-Text Generation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Data-to-Text Generation WebNLG HTML (fine-tuning) BLEU 65.4 #8 of 20 Archive leaderboard report
Data-to-Text Generation WebNLG Full HTLM (prefix 0.1%) BLEU 56.3 #6 of 8 Archive leaderboard report
Table-to-Text Generation DART HTLM (fine-tuning) BERT 0.94 #5 of 6 Archive leaderboard report
Table-to-Text Generation DART HTLM (fine-tuning) BLEU 47.2 #5 of 6 Archive leaderboard report
Table-to-Text Generation DART HTLM (fine-tuning) BLEURT 0.4 #5 of 6 Archive leaderboard report
Table-to-Text Generation DART HTLM (fine-tuning) METEOR 0.39 #5 of 6 Archive leaderboard report
Table-to-Text Generation DART HTLM (fine-tuning) Mover 0.51 #5 of 6 Archive leaderboard report
Table-to-Text Generation DART HTLM (fine-tuning) TER 0.44 #5 of 6 Archive leaderboard report
Table-to-Text Generation DART GPT-2-Large (fine-tuning) BERT 0.94 #6 of 6 Archive leaderboard report
Table-to-Text Generation DART GPT-2-Large (fine-tuning) BLEU 47.0 #6 of 6 Archive leaderboard report
Table-to-Text Generation DART GPT-2-Large (fine-tuning) BLEURT 0.4 #6 of 6 Archive leaderboard report
Table-to-Text Generation DART GPT-2-Large (fine-tuning) METEOR 0.39 #6 of 6 Archive leaderboard report
Table-to-Text Generation DART GPT-2-Large (fine-tuning) Mover 0.51 #6 of 6 Archive leaderboard report
Table-to-Text Generation DART GPT-2-Large (fine-tuning) TER 0.46 #6 of 6 Archive leaderboard report
Table-to-Text Generation E2E HTLM (fine-tuning) BLEU 70.3 #1 of 2 Archive leaderboard report
Table-to-Text Generation E2E HTLM (fine-tuning) CIDEr 2.47 #1 of 2 Archive leaderboard report
Table-to-Text Generation E2E HTLM (fine-tuning) METEOR 46.3 #1 of 2 Archive leaderboard report
Table-to-Text Generation E2E HTLM (fine-tuning) NIST 8.90 #1 of 2 Archive leaderboard report
Table-to-Text Generation E2E HTLM (fine-tuning) ROUGE-L 70.8 #1 of 2 Archive leaderboard report
Table-to-Text Generation E2E GPT-2-Large (fine-tuning) BLEU 68.5 #2 of 2 Archive leaderboard report
Table-to-Text Generation E2E GPT-2-Large (fine-tuning) CIDEr 2.45 #2 of 2 Archive leaderboard report
Table-to-Text Generation E2E GPT-2-Large (fine-tuning) METEOR 46.0 #2 of 2 Archive leaderboard report
Table-to-Text Generation E2E GPT-2-Large (fine-tuning) NIST 8.78 #2 of 2 Archive leaderboard report
Table-to-Text Generation E2E GPT-2-Large (fine-tuning) ROUGE-L 69.9 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (All) HTLM (fine-tuning) BLEU 55.6 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (All) HTLM (fine-tuning) METEOR 0.42 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (All) HTLM (fine-tuning) TER 0.4 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (All) GPT-2-Large (fine-tuning) BLEU 55.5 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (All) GPT-2-Large (fine-tuning) METEOR 0.42 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (All) GPT-2-Large (fine-tuning) TER 0.42 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Seen) HTLM (fine-tuning) BLEU 65.4 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Seen) HTLM (fine-tuning) METEOR 0.46 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Seen) HTLM (fine-tuning) TER 0.33 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Seen) GPT-2-Large (fine-tuning) BLEU 65.3 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Seen) GPT-2-Large (fine-tuning) METEOR 0.46 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Seen) GPT-2-Large (fine-tuning) TER 0.33 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Unseen) HTLM (fine-tuning) BLEU 48.4 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Unseen) HTLM (fine-tuning) METEOR 0.39 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Unseen) HTLM (fine-tuning) TER 0.51 #1 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Unseen) GPT-2-Large (fine-tuning) BLEU 43.1 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Unseen) GPT-2-Large (fine-tuning) METEOR 0.38 #2 of 2 Archive leaderboard report
Table-to-Text Generation WebNLG (Unseen) GPT-2-Large (fine-tuning) TER 0.53 #2 of 2 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections