| --- |
| license: other |
| language: |
| - zh |
| tags: |
| - recipe |
| - rag |
| - embeddings |
| - bge-m3 |
| - android |
| pretty_name: ChiShenMe BGE-M3 Recipe Dataset Packs |
| --- |
| |
| # ChiShenMe BGE-M3 Recipe Dataset Packs |
|
|
| This repository contains offline recipe RAG dataset packs for the Android-first app **吃什么**. |
|
|
| The packs are generated from `recipe_corpus_full.json`, preprocessed into recipe chunks, and embedded with `BAAI/bge-m3` dense vectors. |
|
|
| ## Files |
|
|
| ```text |
| dataset-index.json |
| |
| chishenme-bge-m3-lite-10k/ |
| dataset-pack.json |
| files/ |
| vectors.f32 |
| metadata.jsonl |
| embedding-manifest.json |
| |
| chishenme-bge-m3-medium-100k/ |
| dataset-pack.json |
| files/ |
| vectors.f32 |
| metadata.jsonl |
| embedding-manifest.json |
| |
| chishenme-bge-m3-full/ |
| dataset-pack.json |
| files/ |
| vectors.f32 |
| metadata.jsonl |
| embedding-manifest.json |
| |
| models/ |
| bge-m3-query-onnx/ |
| model-pack.json |
| files/ |
| tokenizer.onnx |
| model.onnx |
| model.onnx_data |
| config.json |
| tokenizer.json |
| sentencepiece.bpe.model |
| ``` |
|
|
| ## Variants |
|
|
| | Level | Chunks | Use case | |
| | --- | ---: | --- | |
| | Lite | 10,000 | Fast Android download and smoke tests | |
| | Medium | 100,000 | Beta testing and better recipe coverage | |
| | Full | 1,552,596 | Full offline recipe retrieval | |
|
|
| ## Android Usage |
|
|
| The app downloads a `dataset-pack.json` file, then downloads the referenced files under `files/`. |
|
|
| Example manifest URL: |
|
|
| ```text |
| https://huggingface.co/datasets/{username}/chishenme-datasets/resolve/main/chishenme-bge-m3-lite-10k/dataset-pack.json |
| ``` |
|
|
| Replace `{username}` with the actual Hugging Face namespace. |
|
|
| The Android ONNX query model pack is available at: |
|
|
| ```text |
| https://huggingface.co/datasets/{username}/chishenme-datasets/resolve/main/models/bge-m3-query-onnx/model-pack.json |
| ``` |
|
|
| ## Format |
|
|
| - `vectors.f32`: contiguous float32 dense vectors, row-major, normalized. |
| - `metadata.jsonl`: one JSON object per vector row. |
| - `embedding-manifest.json`: embedding generation metadata. |
| - `dataset-pack.json`: app-facing pack manifest with file sizes and SHA256 hashes. |
| - `model-pack.json`: app-facing ONNX query model manifest. |
|
|
| ## License |
|
|
| The source dataset is user-provided. Verify redistribution rights before publishing publicly. |
|
|
