Hugging Face's logo

Datasets:
brittlewis12
/
wikiembed.py

Tasks:
Question Answering
Modalities:
Text
Size:
100K<n<1M
Tags:
wikipedia
usearch
distilbert
msmarco
text
License:
Dataset card Files Community
wikiembed.py / README.md
brittlewis12's picture
brittlewis12
Upload folder using huggingface_hub
22b128b verified
preview code
|
raw
history blame contribute delete
4.12 kB
metadata
license: cc-by-sa-3.0
task_categories:
  - question-answering
tags:
  - wikipedia
  - usearch
  - distilbert
  - msmarco
  - text
size_categories:
  - 100K<n<1M

wikiembed.py

text embeddings for wikipedia articles

artifacts

sqlite3 db with article_sections table, containing cleaned wikipedia article contents by section

  • id - unique section id, generated by sqlite
  • article_id - wikipedia’s provided id for the article to which the section belongs
  • title - the article title
  • url - the article url
  • sequence_id - the per-article section order number
  • section_name - the header name from Wikipedia, or 'Lead' which has no heading.
  • text - the cleaned text contents for the section

usearch index for semantic search of wikipedia section contents

  • keyed on the section id from the sqlite article_sections table
  • vector embeddings generated with msmarco distilbert model in coreml
    • 768 dimensions
    • f32 precision
    • 512 token limit
  • section contents chunked & embedded with page title and section name in a chunk header
  • indexed with the following parameters:
    • inner product distance metric
    • connectivity (M in HSNW terms) of 16
      • tested 16, 32, 64, & 200; 16 was smallest disk size with identical recall accuracy on test queries
        • index sizes with various connectivity<>precision config pairs 
          ❯ du -sh ./20240720/connectivity-*/*
979M	./20240720/connectivity-16/simplewiki-20240720.f16.index
1.8G	./20240720/connectivity-16/simplewiki-20240720.f32.index
1.8G	./20240720/connectivity-200/simplewiki-20240720.f16.index
1.8G	./20240720/connectivity-200/simplewiki-20240720.f32.index
1.0G	./20240720/connectivity-32/simplewiki-20240720.f16.index
1.9G	./20240720/connectivity-32/simplewiki-20240720.f32.index
1.2G	./20240720/connectivity-64/simplewiki-20240720.f16.index
2.0G	./20240720/connectivity-64/simplewiki-20240720.f32.index
    • index expansion add (efConstruction in HNSW terms) of 128 (usearch default)
    • quantizing embedding vectors from f32 to f16
      • identical recall accuracy on test queries (i8 performed poorly by contrast)
    • multi key support enabled (so more than one chunk can refer to the same section id)

the original simple english wikipedia dump that was used to generate these artifacts, dated 2024-08-01

modifiable reproduction script wikiembed.py

  • requires msmarco distilbert tas b coreml model as written
  • coremltools prediction only works on macOS*, which was used to generate the vector embeddings for the semantic index. for cross-platform coreml prediction, check out tvm
  • other dependencies described in requirements.txt:
    pip3 install -r requirements.txt
  • make any desired changes to e.g. index parameters, wikipedia dump language or date, or embeddings model in the script
  • run the script to download, clean, persist & index the dump contents
    chmod +x wikiembed.py
./wikiembed.py

released under creative commons license (CC BY-SA 3.0 unported)

data cleaning adapted from olm/wikipedia

by britt lewis