README.md

metadata

license: mit

Benchmark embedding models course

This repository gives you access to large datasets for the embedding models course. Since these files are too big for GitHub, you’ll find them in the datasets folder. Each file includes:

• Text chunks taken from documents.
• Questions that can be answered using those chunks.

Data format

Each JSON file looks like this:

{
    "chunks": [
        {
            "id": 0,
            "text_chunk": "",
            "embeddings": {
                "gemini-embedding-001": [],
            }
        },
        ...
    ],
    "question_answer_pairs": [
        {
            "chunk_id": 0,
            "question": "",
            "embeddings": {
                "gemini-embedding-001": [],
            }
        },
        ...
    ]
}

• Every text_chunk and question has an embeddings field.
• The embeddings field stores dense vectors from different models.

How to use

1. Download this repository, or just the JSON files you need.
2. Put the files in the data/embeddings/ folder of the Benchmark_Embedding_Models repository.
3. You’ll use these files in the following notebooks:
   • 5_1_BenchmarkModelsManually.ipynb
   • 5_2_BenchmarkModelsRanx.ipynb

License

This repository is licensed under the MIT License.