研究者がチェコの HTML ドキュメント処理のためのコンパクトな基盤モデルを導入

一次情報源文書記録されたソース 2026年9月17日 6:33 UTC

出版社: arxiv.org
ソースリンク: arxiv.orghttps://arxiv.org/abs/2609.18494
ソースの種類: 一次文書 — 私たちが直接読む公式発表、論文、提出書類、またはファーストパーティのページ。

コンテキスト60秒で理解できる

ここから始めましょう

重要な用語

基礎モデル: 多くの下流タスクに適応できる大規模な事前トレーニング済みモデル。
分類: モデルが入力を 1 つ以上の事前定義されたカテゴリに割り当てるタスク。
データセット: トレーニング、検証、テストに使用される構造化サンプルまたは非構造化サンプルのコレクション。

何が起こったのか

Researchers have developed a compact called HTML-LM, which is designed to process Czech HTML documents. The model has 154 million parameters and was trained on 100 million web documents using multiple objectives. It sets a new state-of-the-art for and regression applications in the Czech Internet domain.

The model has 154 million parameters and was trained on 100 million web documents using multiple objectives.

It sets a new state-of-the-art for and regression applications in the Czech Internet domain.

ソースの詳細: arxiv.org ↗

なぜそれが重要なのか

The development of HTML-LM is significant because it addresses the limitations of existing approaches to processing web documents. It is a compact model that is both performant and economic, making it suitable for high-traffic industrial environments.

The model is trained on a large of web documents, which allows it to learn structural information inherent in HTML.

It uses a ModernBERT-based architecture, which enables it to process real-world web pages effectively.

The model is deployed in production, processing thousands of web documents per second, making it a practical solution for industrial environments.

It is released to the community under the CC BY-NC 4.0 license, making it available for use by others.

The model sets a new state-of-the-art for and regression applications in the Czech Internet domain, surpassing both larger encoders and small-sized LLMs.

Interactive Mechanism

インタラクティブなメカニズム: 実際にどのように機能するか

この開発の背後にある基盤となるテクノロジーをインタラクティブに探索します。

Thinking Budget (Test-Time Tokens):1,024 tokens

Complex Accuracy79%Math & Code Logic

Latency3.2sTime to first full output

Inference Cost$0.0092Per query estimated

Reasoning StyleStep VerificationInternal chain depth

Active Thinking Trace:

1Deconstruct user problem into formal constraints

2Propose candidate hypotheses & step-by-step calculation

3Self-correction: Backtrack and refute subtle edge cases

4Exhaustive consistency check & final output synthesis

Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.

インタラクティブコンセプトチェック+10 Points

What is AI? Quiz

A route planner searches possible journeys using explicit rules. What does this illustrate about AI?

次に見るべきもの

The development of HTML-LM is a significant step towards creating universal, high-quality representations of web documents in high-traffic industrial environments.

The model's performance and economic viability make it a promising solution for industrial environments.

The model's ability to process real-world web pages effectively makes it a practical solution for a wide range of applications.

The model's release to the community under the CC BY-NC 4.0 license makes it available for use by others.

The model's deployment in production demonstrates its ability to process thousands of web documents per second.

The model's performance in the Czech Internet domain sets a new state-of-the-art, surpassing both larger encoders and small-sized LLMs.