Experiments around APIs for (image/document) AI models.

• CPU only docker setup
• Expects preloaded models, no (annoying) auto-downloads
• Stats about token usages (partially/WIP)

• Image to Data (by donut)
• Visual Document Question Answering (Image) (by donut)
• WIP Document Classification (Image) (by dit)
  • (dataset) RVL-CDIP: "letter", "form", "email", "handwritten", "advertisement", "scientific report", "scientific publication", "specification", "file folder", "news article", "budget", "invoice", "presentation", "questionnaire", "resume", "memo"

general Natural Language Inference

• Question Answering
• Question Answer Generation
• Question Generation
• Question Natural Language Inference / QNLI

Startup server:

docker compose up

• Service Home: localhost:8702
• OpenAPI Docs: localhost:8702/docs (WIP 🚧)

Run CLI in docker container:

# build container before using cli (if never `up`ed before)
docker compose build baigel

# open shell in container:
docker compose run --rm baigel bash

# run cli help:
poetry run cli

# download models:
poetry run cli download

# list models:
poetry run cli models

Manage dependencies with poetry v2:

poetry lock
poetry sync --no-root
poetry install

# poetry lock --regenerate && poetry sync --no-root && poetry install

This project is distributed as free software under the MIT License, see License.

© 2024 Michael Becker https://i-am-digital.eu