ids stringclasses 5
values | texts stringclasses 5
values |
|---|---|
8651ffaa-3b73-44aa-9eee-3f69532f45d0 | Dataset Schema and Field Types: This section specifies the canonical schema for records and the allowed datatypes for each field. Typical fields include id (string, UUIDv4), source_uri (string), created_at (ISO 8601 timestamp), title (short text, <= 120 chars), body (long text), tags (array of strings), and language (I... |
7fbd9b0c-b4e4-4c81-994f-0de2d74772f2 | Governance, Licensing, and PII Policy: This section defines the license for redistribution (for example CC BY 4.0), a provenance log for each record, and a redaction policy for personally identifiable information. PII handling includes deterministic masking for emails and phones, salted hashing for user identifiers, an... |
c3e70ba1-b0f9-452e-b8e8-9b62d5560fb1 | Tokenizer Alignment and Chunk Budget: This section explains how token counts are measured and why chunk budgets are aligned to the target model. For OpenAI- or Llama-family models, tokenization differs, so the tokenizer used during indexing must match the model used at query time. Recommended defaults are chunk_size = ... |
aafab872-d459-4df9-bd27-05f3205f85e3 | Chunking Strategy Matrix: This section documents when to use sentence-aware splitting versus token-only or semantic grouping. Use a sentence-based splitter for general prose where maintaining sentence boundaries increases coherence. Use a token text splitter to enforce strict token budgets regardless of sentence bounda... |
855e86c2-fc30-4d6d-b8a1-674cf09ffb81 | Evaluation Protocols and Dataset Card: This section defines offline retrieval metrics (Recall@k, MRR@k, nDCG@k) and a small, labeled query set for smoke tests. It also specifies a dataset card with sections for motivation, composition, collection process, preprocessing, intended use, limitations, and ethical considerat... |
README.md exists but content is empty.
- Downloads last month
- 7