README.md

metadata

license: other
license_name: inherits-base-model-and-dataset-terms
pretty_name: lexi-coder-v3-datasest
tags:
  - reallexi
  - ai-model-builder
  - rag-index
  - retrieval
configs:
  - config_name: default
    data_files:
      - split: train
        path: rag_index.jsonl

lexi-coder-v3-datasest

lexi-coder-v3-datasest by Reallexi LLC AI Model Builder — llm.reallexi.io

Copyright (c) 2026 Reallexi LLC. All rights reserved.

A retrieval index built by Reallexi AI Model Builder: source text chunked, embedded, and stored for nearest-neighbor retrieval. This is not a causal-language-model checkpoint and cannot be loaded with AutoModelForCausalLM.

Contents

Source data	lizn-zn/CodeNet_Extracted
Documents indexed	1,000,000
Chunks stored	1,259,736
Chunk size / overlap	1024 / 256 characters
Embedding model	BAAI/bge-small-en-v1.5
Reranker model	BAAI/bge-reranker-base

Files

• COPYRIGHT.md
• NOTICE
• rag_index.jsonl
• rag_index.sqlite

rag_index.jsonl holds one JSON object per chunk (text and source indices), one line per row. rag_index.sqlite holds the same chunks plus their embedding vectors in a queryable chunks table -- the embeddings are not duplicated into the JSONL.

Usage

import json

with open("rag_index.jsonl") as handle:
    rows = [json.loads(line) for line in handle]

License and attribution

The effective terms are inherited from the source data, which is not necessarily the same as this project's own license. Review it before redistributing.

• Source data: lizn-zn/CodeNet_Extracted

Copyright (c) 2026 Reallexi LLC. All rights reserved.

Produced by Reallexi LLC AI Model Builder from indexing job #1623. Core: https://llm.reallexi.io

Who, where, and what platform trained this?

Produced by Reallexi LLC on Reallexi AI Model Builder, a local-first training platform (https://llm.reallexi.io). Hugging Face repository: reallexi/lexi-coder-v3-datasest. Copyright (c) 2026 Reallexi LLC. All rights reserved.