leo20000306/KANT_GoogleBing_Corpus

KANT-8 Google RAG Corpus

This folder is an HF-ready text-only release of the KANT-8 training JSONs with Google/Serper search results embedded into each original sample. Images are not copied into this release; each record keeps the original image field as text.

Each row preserves the original training JSON fields and adds:

• —dataset: KANT dataset name.
• —source_index: index in the original trainjson/*_train.json file.
• —source_json: local source JSON path used to build the row.
• —entities: query-oriented entities used for search.
• —serper_query: first Google/Serper query for the row, when available.
• —search_queries: structured query metadata.
• —references: up to five Google/Serper search references with title, time, url, content, rank, serper_raw, search_entity, search_query, search_engine, serper_endpoint, and serper_params when available.

Fields that expose retrieval time or redundant source labeling, including retrieved_at, source_type, and content_extended, are intentionally removed.

Files

• —data/by_dataset/*.jsonl: one JSONL shard per KANT-8 dataset.
• —manifest.json: generation metadata and record counts.

Summary

Total records: 584923

Total references: 2551798