#### NOTE
This is a **Hugging Face dataset**. For large datasets, ensure `huggingface_hub>=1.1.3` to avoid rate limits. Learn more in the Hugging Face integration docs.
# FiftyOne Embeddings Dataset
This dataset combines the FiftyOne Q&A and function calling datasets with pre-computed embeddings for fast similarity search.
## Dataset Information
- **Total samples**: 28,118
- **Q&A samples**: 14,069
- **Function samples**: 14,049
- **Embedding model**: text-embedding-3-large
- **Embedding dimension**: 3072
## Schema
- `query`: The original question/query text
- `response`: The unified response content (either answer text for Q&A or function call text for function samples)
- `content_type`: Either ‘qa_response’ or ‘function_call’
- `embedding`: Pre-computed embedding vector (3072 dimensions), based on the `query` feature
- `dataset_type`: Either ‘qa’ or ‘function’
- `source_dataset`: Original dataset name
- `embedding_model`: Model used to compute embeddings
## Usage
```python
from datasets import load_dataset
import numpy as np
from scipy.spatial.distance import cosine
# Load dataset with embeddings
dataset = load_dataset("Voxel51/fiftyone-embeddings-combined", split="train")
# Extract embeddings for similarity search
embeddings = np.array([item['embedding'] for item in dataset])
queries = [item['query'] for item in dataset]
def find_similar(query_embedding, top_k=5):
similarities = [1 - cosine(query_embedding, emb) for emb in embeddings]
top_indices = np.argsort(similarities)[-top_k:][::-1]
results = []
for i in top_indices:
item = dataset[i]
results.append({
'query': item['query'],
'response': item['response'], # Unified response field
'type': item['content_type'],
'similarity': similarities[i]
})
return results
```