FineWeb Edu: 1.5 Billion Educational Web Passages with Embeddings

Over 1.5 billion rows of educational web passages filtered from the FineWeb dataset. Each passage includes cleaned text, metadata, and 384-dimensional text embeddings. The dataset is provided by lance-format in the Lance format, optimized for retrieval-heavy AI workloads.

Use Cases

• Train or benchmark text retrieval models based on the 384-dim embeddings.
• Build educational question-answering systems based on cleaned web passages.
• Analyze the content and structure of educational web pages based on the metadata.
• Create semantic search indexes for educational content based on the text embeddings.

Strengths

• Over 1.5 billion rows of educational web content.
• Includes 384-dimensional text embeddings for each passage.
• Text is cleaned and accompanied by metadata.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-02-02 14:20:05; freshness should be verified.

Provenance

Source

Filtered from the FineWeb dataset.

Collection Method

Likely filtered from web crawl data.

Freshness

2026-02-02 14:20:05