---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: text
    dtype: string
  - name: url
    dtype: string
  - name: date
    dtype: timestamp[ns, tz=UTC]
  - name: dump
    dtype: string
  - name: file_path
    dtype: string
  - name: language_score
    dtype: float64
  - name: minhash_cluster_size
    dtype: int64
  - name: top_langs
    dtype: string
  - name: score
    dtype: float64
  - name: int_score
    dtype: int64
  - name: topic_class_1
    dtype: string
  - name: topic_prob_1
    dtype: float64
  - name: topic_class_2
    dtype: string
  - name: topic_prob_2
    dtype: float64
  - name: topic_class_3
    dtype: string
  - name: topic_prob_3
    dtype: float64
  - name: format_class_1
    dtype: string
  - name: format_prob_1
    dtype: float64
  - name: format_class_2
    dtype: string
  - name: format_prob_2
    dtype: float64
  - name: format_class_3
    dtype: string
  - name: format_prob_3
    dtype: float64
  - name: age_group_class_1
    dtype: string
  - name: age_group_prob_1
    dtype: float64
  - name: age_group_class_2
    dtype: string
  - name: age_group_prob_2
    dtype: float64
  - name: age_group_class_3
    dtype: string
  - name: age_group_prob_3
    dtype: float64
  splits:
  - name: train
    num_bytes: 225675034950
    num_examples: 54128784
  download_size: 131804901421
  dataset_size: 225675034950
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# FineWeb2-Ro-BERT

**FineWeb2-Ro-BERT** is a large-scale pretraining dataset in the Romanian language. The data is derived from [FineWeb2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2) and annotated using a bert architecture for signals such as `educational quality` or `topic`. More details can be found [here](https://arxiv.org/abs/2511.01090).

## Key Features

* **Massive Scale**: Contains approximately **54.1M** rows (documents or sequences), providing comprehensive linguistic coverage for training robust Romanian embeddings and encoders.

## Usage

You can load this dataset using the Hugging Face `datasets` library:

```python
from datasets import load_dataset

dataset = load_dataset("OpenLLM-Ro/fineweb2-ro-bert", split="train")