|
|
--- |
|
|
title: Grounding Snippet Generator |
|
|
emoji: 🚀 |
|
|
colorFrom: red |
|
|
colorTo: red |
|
|
sdk: docker |
|
|
app_port: 8501 |
|
|
tags: |
|
|
- streamlit |
|
|
pinned: false |
|
|
short_description: Streamlit template space |
|
|
--- |
|
|
|
|
|
--- |
|
|
title: Snippet Generator |
|
|
emoji: ✂️ |
|
|
colorFrom: blue |
|
|
colorTo: purple |
|
|
sdk: streamlit |
|
|
sdk_version: 1.28.0 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
# ✂️ Snippet Generator |
|
|
|
|
|
Recreates Google Vertex AI / Gemini grounding-style extractive snippets. |
|
|
|
|
|
## How it works |
|
|
|
|
|
1. **Sentence Segmentation** - Splits document on sentence boundaries and newlines, filters noise (URLs, questions, low-alpha content) |
|
|
2. **Cross-Encoder Scoring** - Uses `cross-encoder/ms-marco-electra-base` to score query-sentence relevance |
|
|
3. **Budget Selection** - Picks top-scoring sentences within character/count limits |
|
|
4. **Document-Order Stitching** - Reassembles in original order with `...` for gaps |
|
|
|
|
|
## Model |
|
|
|
|
|
Uses [MS MARCO Cross-Encoder](https://huggingface.co/cross-encoder/ms-marco-electra-base) trained on search relevance data - the same task as snippet generation. |
|
|
|
|
|
## Usage |
|
|
|
|
|
1. Enter a search query |
|
|
2. Paste document content |
|
|
3. Adjust settings (max chars, max sentences) |
|
|
4. Click "Generate Snippet" |
|
|
|
|
|
## Example |
|
|
|
|
|
**Query:** `best prostate cancer treatment in the world` |
|
|
|
|
|
**Output:** |
|
|
``` |
|
|
This makes Asklepios one of the best prostate cancer treatment centers for foreigners. ... Spain is a leader in prostate cancer treatment, with top clinics like Centro Médico Teknon, Quironsalud Madrid, and Hospital Quiron Barcelona. ... |
|
|
``` |
