OmniAlign

<p align="center"> <a href="https://huggingface.co/WPS-Qingqiu/OmniAlign/blob/main/README_CN.md">中文</a>&nbsp;｜&nbsp;English </p> <p align="center"> <a href="https://arxiv.org/abs/2608.18474">📄 Paper (arXiv)</a>&nbsp;|&nbsp; <a href="https://github.com/MilkDargon/OmniAlign">💻 GitHub</a>&nbsp;|&nbsp; <a href="https://huggingface.co/spaces/hugging-apps/omnialign-demo">🕹️ Demo</a> </p>

OmniAlign

Universal multilingual sequence alignment model for sentence alignment and word alignment.

OmniAlign is a universal multilingual alignment model designed to deliver efficient sentence alignment and word alignment through multi-stage training. It combines contextual word embeddings for token-level alignment with sentence embeddings for cross-lingual semantic matching, and progressively strengthens both through a four-stage training pipeline (pretrain → unsupervised → supervised → distillation).

• Word alignment: infers token similarity from contextual word embeddings and derives subword-level alignments.
• Sentence alignment: encodes source and target sentences into a shared vector space, then matches them with a dynamic-programming two-step approach — first pinning down approximate one-to-one anchors, then searching for all valid alignments under anchor constraints.
• Four-stage training: progressively strengthens both word and sentence embeddings through four training stages.

This model is the final checkpoint produced by the four-stage distillation pipeline, built on top of Alibaba-NLP/gte-multilingual-mlm-base.

This repository hosts model weights and configs only.
Inference code, alignment scripts, and runnable examples are in the GitHub repo: MilkDargon/OmniAlign.
Loading the weights here gives you the encoder; word/sentence alignment is implemented in the GitHub example/ code.

---

Demo

An interactive demo is available on Hugging Face Spaces: omnialign-demo.

It runs on free ZeroGPU and is built on this repository. We thank the Hugging Face open-source team for creating this Space so others can try OmniAlign in the browser.

---

Model Details

---

Installation

Clone the GitHub repo (code), then install inference dependencies. Python 3.12 is recommended.

git clone https://github.com/MilkDargon/OmniAlign.git
cd OmniAlign
hf download WPS-Qingqiu/OmniAlign --local-dir OmniAlign
pip install -r example/requirements.txt

The example needs only inference dependencies (torch, transformers, sentence_transformers, jieba, nltk, sentence_splitter, numpy, numba, faiss-cpu).

Weights are downloaded from this Hub repo into a local OmniAlign/ folder. If the hf command is missing, run pip install -U huggingface_hub first.

---

Quick Start

From the cloned GitHub repo, set --model_path to the downloaded weights directory:

Word Alignment

python example/example.py \
  --src "他没有遵守承诺" \
  --tgt "He broke promise" \
  --src_lang zh --tgt_lang en \
  --task word_align \
  --model_path OmniAlign

Output:

===Word Alignment Result ===
他  <==>  He
没有  <==>  broke
遵守  <==>  broke
承诺  <==>  promise

Sentence Alignment

python example/example.py \
  --src "Large language models (LLMs) represent one of the most transformative breakthroughs in artificial intelligence over the past decade. Trained on terabytes of text data spanning books, websites, and academic papers, these models excel at capturing complex linguistic patterns, contextual nuances, and even logical reasoning." \
  --tgt "大语言模型（LLMs）是过去十年人工智能领域最具变革性的突破之一。这些模型基于涵盖书籍、网站与学术论文在内的万亿字节级文本数据进行训练，擅长捕捉复杂的语言模式、语境细微差异乃至逻辑推理能力。" \
  --src_lang en --tgt_lang zh \
  --task sent_align \
  --model_path OmniAlign

Output:

===Sentence Alignment Result ===
Large language models (LLMs) represent one of the most transformative breakthroughs in artificial intelligence over the past decade.  <==>  大语言模型（LLMs）是过去十年人工智能领域最具变革性的突破之一。
Trained on terabytes of text data spanning books, websites, and academic papers, these models excel at capturing complex linguistic patterns, contextual nuances, and even logical reasoning.  <==>  这些模型基于涵盖书籍、网站与学术论文在内的万亿字节级文本数据进行训练，擅长捕捉复杂的语言模式、语境细微差异乃至逻辑推理能力。

See example/README.md in the GitHub repo for the full list of arguments.

---

Evaluation

Word Alignment (AER, lower is better)

AER (Alignment Error Rate) on standard word alignment test sets. For each language pair, the best result is <span style="color:#d1242f"><strong>red and bold</strong></span>, and the second-best is <span style="color:#0969da"><u>blue and underlined</u></span>.

<div translate="no"> <table> <thead> <tr> <th>Methods</th> <th>zh-en</th> <th>de-en</th> <th>fr-en</th> <th>ro-en</th> <th>ja-en</th> <th>es-en</th> <th>pt-en</th> <th>ru-en</th> <th>it-en</th> </tr> </thead> <tbody> <tr> <td>FastAlign</td> <td>38.1</td><td>27.0</td><td>10.5</td><td>27.0</td><td>51.1</td><td>-</td><td>-</td><td>-</td><td>-</td> </tr> <tr> <td>GIZA++</td> <td>35.1</td><td>20.6</td><td>5.9</td><td>26.4</td><td>48.0</td><td>-</td><td>-</td><td>-</td><td>-</td> </tr> <tr> <td>SimAlign</td> <td>21.6</td><td>16.6</td><td>7.5</td><td>22.3</td><td>46.6</td><td>14.2</td><td>14.1</td><td>15.4</td><td>17.7</td> </tr> <tr> <td>AwesomeAlign</td> <td>13.3</td><td>13.3</td><td>3.8</td><td>18.7</td><td>37.4</td><td>12.0</td><td>12.7</td><td><span style="color:#0969da"><u>13.5</u></span></td><td>15.7</td> </tr> <tr> <td>AccAlign</td> <td>11.5</td><td>12.1</td><td>2.8</td><td>16.9</td><td>36.8</td><td><span style="color:#0969da"><u>11.1</u></span></td><td><span style="color:#0969da"><u>12.1</u></span></td><td>12.5</td><td><span style="color:#0969da"><u>14.3</u></span></td> </tr> <tr> <td>WSPAlign (Bilingual)</td> <td>13.1</td><td>11.1</td><td>2.8</td><td><span style="color:#0969da"><u>10.1</u></span></td><td><span style="color:#0969da"><u>19.3</u></span></td><td>-</td><td>-</td><td>-</td><td>-</td> </tr> <tr> <td>WSPAlign (Multilingual)</td> <td>22.3</td><td>20.0</td><td>12.8</td><td>26.4</td><td>45.8</td><td>13.4</td><td>12.3</td><td>13.1</td><td>17.1</td> </tr> <tr> <td>BinaryAlign (Bilingual)</td> <td><span style="color:#d1242f"><strong>4.8</strong></span></td><td><span style="color:#d1242f"><strong>7.8</strong></span></td><td><span style="color:#d1242f"><strong>1.9</strong></span></td><td><span style="color:#d1242f"><strong>7.4</strong></span></td><td><span style="color:#d1242f"><strong>14.3</strong></span></td><td>-</td><td>-</td><td>-</td><td>-</td> </tr> <tr> <td><strong>OmniAlign (ours)</strong></td> <td><span style="color:#0969da"><u>8.5</u></span></td><td><span style="color:#0969da"><u>11.0</u></span></td><td><span style="color:#0969da"><u>2.7</u></span></td><td>16.7</td><td>29.6</td><td><span style="color:#d1242f"><strong>10.7</strong></span></td><td><span style="color:#d1242f"><strong>11.9</strong></span></td><td><span style="color:#d1242f"><strong>12.1</strong></span></td><td><span style="color:#d1242f"><strong>14.1</strong></span></td> </tr> </tbody> </table> </div>

Sentence Alignment (F1, higher is better)

F1 scores on standard sentence alignment test sets. For each language pair, the best result is <span style="color:#d1242f"><strong>red and bold</strong></span>, and the second-best is <span style="color:#0969da"><u>blue and underlined</u></span>.

<div translate="no"> <table> <thead> <tr> <th>Algorithm</th> <th>en-zh</th> <th>en-es</th> <th>en-it</th> <th>en-de</th> <th>en-fr</th> <th>en-ru</th> <th>de-fr</th> </tr> </thead> <tbody> <tr> <td>Gale–Church</td> <td>0.682</td><td>0.900</td><td>0.977</td><td>0.897</td><td>0.838</td><td>0.911</td><td>0.680</td> </tr> <tr> <td>BleuAlign</td> <td>0.782</td><td>0.819</td><td>0.901</td><td>0.806</td><td>0.757</td><td>0.791</td><td>0.770</td> </tr> <tr> <td>VecAlign</td> <td>0.957</td><td>0.892</td><td>0.956</td><td>0.869</td><td>0.880</td><td>0.921</td><td>0.902</td> </tr> <tr> <td>BertAlign</td> <td><span style="color:#0969da"><u>0.969</u></span></td><td><span style="color:#0969da"><u>0.897</u></span></td><td><span style="color:#d1242f"><strong>0.984</strong></span></td><td><span style="color:#0969da"><u>0.900</u></span></td><td><span style="color:#0969da"><u>0.909</u></span></td><td><span style="color:#d1242f"><strong>0.938</strong></span></td><td><span style="color:#d1242f"><strong>0.939</strong></span></td> </tr> <tr> <td>SentAlign</td> <td>0.968</td><td>0.872</td><td>0.978</td><td>0.892</td><td>0.903</td><td>0.920</td><td><span style="color:#0969da"><u>0.932</u></span></td> </tr> <tr> <td>CrocoAlign</td> <td>0.660</td><td>0.696</td><td>0.864</td><td>0.804</td><td>0.788</td><td>0.783</td><td>0.714</td> </tr> <tr> <td><strong>OmniAlign (ours)</strong></td> <td><span style="color:#d1242f"><strong>0.970</strong></span></td><td><span style="color:#d1242f"><strong>0.906</strong></span></td><td><span style="color:#0969da"><u>0.978</u></span></td><td><span style="color:#d1242f"><strong>0.913</strong></span></td><td><span style="color:#d1242f"><strong>0.912</strong></span></td><td><span style="color:#0969da"><u>0.935</u></span></td><td>0.922</td> </tr> </tbody> </table> </div>

---

Citation

@misc{yang2026omnialign,
  title         = {OmniAlign: A Unified Multilingual Aligner for Word and Sentence Alignment},
  author        = {Mengpeng Yang and Jingxu Yang and Chao Chen and Tian Xia and Yabo Sun and Qiang Liu},
  year          = {2026},
  eprint        = {2608.18474},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2608.18474}
}

---

License

This model is released under the Apache License, Version 2.0.

Read this in 中文.