| --- |
| license: apache-2.0 |
| language: |
| - fr |
| - en |
| - es |
| task_categories: |
| - visual-document-retrieval |
| - text-retrieval |
| tags: |
| - RAG |
| - DSE |
| - retrieval |
| - energy |
| configs: |
| - config_name: train |
| data_files: "train-*.parquet" |
| - config_name: filtered |
| data_files: "output/filtered-*.parquet" |
| --- |
| |
|
|
|
|
|
|
| # VDR - Organized, Grouped, Cleaned |
| # Energy Vision DSE |
|
|
| > **Intended for image/text to vector (DSE)** |
|
|
| ## Dataset Composition |
|
|
| Made with https://github.com/RacineAIOS/VDR_pdf-to-parquet |
| |
| This dataset was created by scraping PDF documents from online sources and generating relevant synthetic queries. |
| We used Google's Gemini 2.0 Flash Lite model in our custom pipeline to produce the queries, allowing us to create a diverse set of questions based on the document content. |
| |
| ## Dataset Statistics |
| |
| | Metric | Value | |
| |--------|-------| |
| | Train rows | 143k | |
| | Test rows | 16.4k | |
| |
| ## Language Distribution |
| |
| | Language | Ratio (%) | |
| |--------|-------| |
| | English (en) | 73.3 | |
| | French (fr) | 26.4 | |
| | Spanish (es) | 0.3 | |
| |
| |
| ## Creators |
| |
| Dataset curated by: |
| - **Paul Lemaistre** |
| - **Léo Appourchaux** |
