---
license: cc-by-sa-4.0
language:
- multilingual
- tr
- en
task_categories:
- sentence-similarity
- feature-extraction
tags:
- embeddings
- distillation
- retrieval
- wikipedia
- embeddinggemma
pretty_name: Wikipedia 40 Languages with EmbeddingGemma-2 Retrieval Embeddings
source_datasets:
- alibayram/wikipedia-40-langs
dataset_info:
  features:
  - name: lang
    dtype: string
  - name: title
    dtype: string
  - name: text
    dtype: string
  - name: url
    dtype: string
  - name: truncated
    dtype: bool
  - name: original_text_chars
    dtype: int32
  - name: embedgemma2_query_embedding
    list: float32
    length: 768
  - name: embedgemma2_document_embedding
    list: float32
    length: 768
  splits:
  - name: test
    num_bytes: 629625358
    num_examples: 58000
  - name: validation
    num_bytes: 1889874358
    num_examples: 174000
  - name: train
    num_bytes: 6706701076
    num_examples: 580000
  download_size: 6180693596
  dataset_size: 8596575434
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
  - split: validation
    path: data/validation-*
  - split: train
    path: data/train-*
---

# Wikipedia 40 Languages with EmbeddingGemma-2 Retrieval Embeddings

Precomputed teacher embeddings from [google/embeddinggemma-2](https://huggingface.co/google/embeddinggemma-2)
for offline embedding distillation, built on [alibayram/wikipedia-40-langs](https://huggingface.co/datasets/alibayram/wikipedia-40-langs).
It is the distillation corpus for [alibayram/embeddingmagibu2](https://huggingface.co/alibayram/embeddingmagibu2),
following the pipeline of [arXiv:2605.29992](https://arxiv.org/abs/2605.29992).

> **Status:** splits are added as they finish (test → validation → train).
> Live progress and ETA: [`progress.json`](./progress.json). Files under `progress/` are partial
> chunks of the split being encoded and are removed once that split is pushed.

## Splits

| Split | Rows |
|---|---|
| test | 58,000 |
| validation | 174,000 |
| train | 580,000 |

## Columns

| Column | Description |
|---|---|
| `lang`, `title`, `url` | From the source dataset |
| `text` | Article text, **clipped** so the document fits 8,192 tokens in both tokenizers (see below). This is exactly the text that was embedded. |
| `truncated` | `true` if `text` was clipped (~2.6% of rows) |
| `original_text_chars` | Character length of the unclipped article |
| `embedgemma2_query_embedding` | Teacher embedding of `task: search result \| query: {title}` (768-d, L2-normalized) |
| `embedgemma2_document_embedding` | Teacher embedding of `title: {title} \| text: {text}` (768-d, L2-normalized) |

Both targets use EmbeddingGemma's retrieval prompt formats, so a student can learn query and
document representations; the title query and its article form a natural positive pair.

## Clipping to both tokenizers

The student model uses the Turkish-optimized [magibu/Altus](https://huggingface.co/magibu/Altus)
tokenizer, which needs ~21% fewer tokens for Turkish but up to ~15% more for some languages
(e.g. Russian). To guarantee that student and teacher always see the **same** input, each rendered
document (`title: … | text: …`) was clipped at the first character where *either* the teacher
tokenizer or Altus exceeds 8,184 tokens (8,192 minus a reserve for special tokens), preferring a
word boundary. No row exceeds 8,192 tokens in either tokenizer.

## How it was made

- Teacher: `google/embeddinggemma-2`, bf16 on a single H200, mean pooling, normalized outputs.
  Verified against fp32 `SentenceTransformer.encode` (cosine ≥ 0.9997).
- Global length sorting with token-budget batching; script:
  `prepare_distillation_dataset.py`.

## Usage

```python
from datasets import load_dataset

ds = load_dataset("alibayram/wikipedia-40-langs-with-embeddings-embeddinggemma2", split="test")
row = ds[0]
doc_input = f"title: {row['title']} | text: {row['text']}"   # student input for the document target
query_input = f"task: search result | query: {row['title']}"  # student input for the query target
```

## License

Text is from Wikipedia under CC BY-SA 4.0; embeddings are derived from it.