Datasets:
File size: 15,022 Bytes
01f6e26 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 | ---
configs:
- config_name: SyntheticClassificationData
data_files:
- path: SyntheticClassificationData/queries.parquet
split: train
- config_name: SyntheticClassificationData_corpus
data_files:
- path: SyntheticClassificationData/corpus.parquet
split: train
- config_name: HotpotQA
data_files:
- path: HotpotQA/queries.parquet
split: train
- config_name: HotpotQA_corpus
data_files:
- path: HotpotQA/corpus.parquet
split: train
- config_name: FEVER
data_files:
- path: FEVER/queries.parquet
split: train
- config_name: FEVER_corpus
data_files:
- path: FEVER/corpus.parquet
split: train
- config_name: MAmmoTH2
data_files:
- path: MAmmoTH2/queries.parquet
split: train
- config_name: MAmmoTH2_corpus
data_files:
- path: MAmmoTH2/corpus.parquet
split: train
- config_name: MIRACL
data_files:
- path: MIRACL/queries.parquet
split: train
- config_name: MIRACL_corpus
data_files:
- path: MIRACL/corpus.parquet
split: train
- config_name: NaturalQuestions
data_files:
- path: NaturalQuestions/queries.parquet
split: train
- config_name: NaturalQuestions_corpus
data_files:
- path: NaturalQuestions/corpus.parquet
split: train
- config_name: NFCorpus
data_files:
- path: NFCorpus/queries.parquet
split: train
- config_name: NFCorpus_corpus
data_files:
- path: NFCorpus/corpus.parquet
split: train
- config_name: PAQ
data_files:
- path: PAQ/queries.parquet
split: train
- config_name: PAQ_corpus
data_files:
- path: PAQ/corpus.parquet
split: train
- config_name: SciFact
data_files:
- path: SciFact/queries.parquet
split: train
- config_name: SciFact_corpus
data_files:
- path: SciFact/corpus.parquet
split: train
- config_name: SQuAD
data_files:
- path: SQuAD/queries.parquet
split: train
- config_name: SQuAD_corpus
data_files:
- path: SQuAD/corpus.parquet
split: train
- config_name: GooAQ
data_files:
- path: GooAQ/queries.parquet
split: train
- config_name: GooAQ_corpus
data_files:
- path: GooAQ/corpus.parquet
split: train
- config_name: TriviaQA
data_files:
- path: TriviaQA/queries.parquet
split: train
- config_name: TriviaQA_corpus
data_files:
- path: TriviaQA/corpus.parquet
split: train
- config_name: MSMARCO
data_files:
- path: MSMARCO/queries.parquet
split: train
- config_name: MSMARCO_corpus
data_files:
- path: MSMARCO/corpus.parquet
split: train
- config_name: EmotionClassification
data_files:
- path: EmotionClassification/queries.parquet
split: train
- config_name: EmotionClassification_corpus
data_files:
- path: EmotionClassification/corpus.parquet
split: train
task_categories:
- text-retrieval
- text-ranking
- sentence-similarity
- text-classification
language:
- multilingual
---
# Embed Nemotron Dataset V1
## Versions
| Date | Commit | Changes |
|-------------|--------------|----------|
| **2026-01-05** | [8808454](https://huggingface.co/datasets/nvidia/embed-nemotron-dataset-v1/tree/880845412bbb3c160940666a4532c6ee9d50f41d) | Initial Release |
## Dataset Description
This dataset is a compilation of high quality fine-tuning datasets that support NVIDIA's release of [llama-embed-nemotron-8b](https://huggingface.co/nvidia/llama-embed-nemotron-8b) model.
[llama-embed-nemotron-8b](https://huggingface.co/nvidia/llama-embed-nemotron-8b) is a versatile text embedding model trained by NVIDIA and optimized for retrieval, reranking, semantic similarity, and classification use cases. This model has robust capabilities for multilingual and cross-lingual text retrieval and is designed to serve as a foundational component in text-based Retrieval-Augmented Generation (RAG) systems. This model achieves state-of-the-art performance on the multilingual [MTEB](https://huggingface.co/spaces/mteb/leaderboard) leaderboard as of October 21, 2025.
`embed-nemotron-dataset-v1` is a selected subset of the fine-tuning data used for training the [llama-embed-nemotron-8b](https://huggingface.co/nvidia/llama-embed-nemotron-8b) model. This data mix is intentionally curated from a variety of sources to ensure robustness and high performance across a broad range of tasks. It comprises both non-synthetic data drawn from established public datasets like MIRACL, HotpotQA, MS MARCO etc.; and synthetic data specifically generated to enhance diversity in task types (for example, classification). Each dataset is additionally transformed with a sophisticated hard negative mining process to maximize the learning signal. Read more details in our technical report: [Llama-Embed-Nemotron-8B: A Universal Text Embedding Model for Multilingual and Cross-Lingual Tasks](https://arxiv.org/abs/2511.07025).
In the table below, you can see all the subdatasets that we are publishing, with their details. In some cases, the data is redistributed directly. For the remaining ones, we only provide IDs of the queries and documents. In the latter case, we provide [python data processing script](https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/biencoder/llama_embed_nemotron_8b/data_preparation.py) to fetch original texts given provided IDs.
This dataset is for research and development only.
---
## Dataset Owner
NVIDIA Corporation
---
## Dataset Creation Date
October 21, 2025
---
## License/Terms of Use
**Dataset Governing Terms**:
Use of each dataset in this collection is governed by the license accompanying it, except for the `FEVER` and `MIRACL` datasets, which are governed by the [Creative Commons Attribution-ShareAlike 4.0 International License](https://creativecommons.org/licenses/by-sa/4.0/) (CC BY-SA 4.0), and the `SyntheticClassificationData`, which is governed by the [Creative Commons Attribution 4.0 International License](https://creativecommons.org/licenses/by/4.0/) (CC BY 4.0).
**Dataset Additional Information**:
The `SyntheticClassificationData` dataset contains synthetic data created using the following models: `meta-llama/Llama-3.3-70B-Instruct`, `meta-llama/Llama-4-Scout-17B-16E-Instruct` and `meta-llama/Llama-4-Maverick-17B-128E-Instruct`. If this dataset is used to create, train, fine tune, or otherwise improve an AI model, which is distributed or made available, such AI model may be subject to redistribution and use requirements in the [Llama 3.3 Community License Agreement](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct/blob/main/LICENSE) and [Llama 4 Community License Agreement](https://www.llama.com/llama4/license/).
---
## Intended Usage
The `embed-nemotron-dataset-v1` is intended to be used by the community to continue to improve open models. The data may be freely used to train and evaluate.
---
## Dataset Composition
| Dataset Name | Task | Samples (Queries / Documents) | Size (MB) | Governing Terms | Source |
|---------------------------------------|----------------|----------:|----------:|-----------------|--------|
| [EmotionClassification](./EmotionClassification/README.md) | Classification | 13,039 / 6 | 0.2 | | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [FEVER](./FEVER/README.md) | Retrieval | 140,085 / 235,948 | 118.2 | CC BY-SA 4.0 | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [GooAQ](./GooAQ/README.md) | Retrieval | 100,000 / 86,393 | 5.5 | | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [HotpotQA](./HotpotQA/README.md) | Retrieval | 170,000 / 754,756 | 21.4 | | <span style="display-inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [MAmmoTH2](./MAmmoTH2/README.md) | Retrieval | 317,180 / 317,205 | 27.8 | | <span style="display-inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [MIRACL](./MIRACL/README.md) | Retrieval | 79,648 / 2,571,803 | 1,480.0 | CC BY-SA 4.0 | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [MSMARCO](./MSMARCO/README.md) | Retrieval | 532,751 / 3,672,883 | 80.6 | | <span style="display-inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [NFCorpus](./NFCorpus/README.md) | Retrieval | 3,685 / 3,573 | 0.2 | | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [NaturalQuestions](./NaturalQuestions/README.md) | Retrieval | 100,231 / 75,215 | 5.1 | | <span style="display-inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [PAQ](./PAQ/README.md) | Retrieval | 1,000,000 / 932,307 | 108.0 | | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [SQuAD](./SQuAD/README.md) | Retrieval | 87,599 / 18,891 | 3.4 | | <span style="display-inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [SciFact](./SciFact/README.md) | Retrieval | 919 / 3,255 | 0.1 | | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| [SyntheticClassificationData](./SyntheticClassificationData/README.md) | Classification | 1,044,212 / 382,227 | 440.1 | CC BY 4.0 | <span style="display-inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#C9D6FF;white-space:nowrap">synthetic</span> |
| [TriviaQA](./TriviaQA/README.md) | Retrieval | 73,346 / 64,137 | 3.8 | | <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span> |
| **Total** (14) | | 3,662,695 / 9,118,599 | 2,314.4 | | | |
## Tag Legend
* <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#CFEBD9;white-space:nowrap">public</span>: Origin of the data is another public dataset
* <span style="display:inline-block;padding:.15em .55em;border-radius:.5em;font:600 12px/1.25 system-ui,-apple-system,Segoe UI,Roboto,Ubuntu,Cantarell,Noto Sans,sans-serif;color:#444;background:#C9D6FF;white-space:nowrap">synthetic</span>: The data was synthetically generated
---
## Dataset Quantification
- **Total Number of Datasets**: 14
- **Total Number of Samples**: 3,662,695 Queries; 9,118,599 Documents
- **Total Size**: 2.3 GB
---
## Dataset Characterization
### **Data Collection Method**
Hybrid: Synthetic, Automated, Human
### **Labeling Method**
Hybrid: Synthetic, Automated, Human
---
## Dataset Format
Each given dataset includes two parts:
1. **Queries**
- `question_id` -- id of the query
- `question` -- query text (or query ID if texts are not redistributed directly)
- `corpus_id` -- name of the dataset
- `pos_doc` -- list of the positive documents for the given query
- `neg_doc` -- list of the negative documents for the given query
2. **Corpus**
- `id` -- document id corresponding to `pos_doc` and `neg_doc` lists from the Queries
- Optional `text` -- document text if texts are redistributed directly
---
## Usage
The data has been prepared to be used with [NeMo AutoModel](https://github.com/NVIDIA-NeMo/Automodel) framework.
This dataset is a compilation of high quality fine-tuning datasets that support NVIDIA's release of [llama-embed-nemotron-8b](https://huggingface.co/nvidia/llama-embed-nemotron-8b) model. To prepare the data and start training the model, follow the steps provided [here](https://github.com/NVIDIA-NeMo/Automodel/tree/main/examples/biencoder/llama_embed_nemotron_8b).
---
## Citation:
```
@misc{babakhin2025llamaembednemotron8buniversaltextembedding,
title={Llama-Embed-Nemotron-8B: A Universal Text Embedding Model for Multilingual and Cross-Lingual Tasks},
author={Yauhen Babakhin and Radek Osmulski and Ronay Ak and Gabriel Moreira and Mengyao Xu and Benedikt Schifferer and Bo Liu and Even Oldridge},
year={2025},
eprint={2511.07025},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2511.07025},
```
---
## Ethical Considerations:
NVIDIA believes **Trustworthy AI** is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications.
When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report quality, risk, security vulnerabilities or **NVIDIA AI Concerns** [here](https://app.intigriti.com/programs/nvidia/nvidiavdp/detail).
|