Talk2Book 📖

Use AI to personify books, so that you can talk to them 🙊

Notebooks

Talk2Book.ipynb

Use this to talk to '1984'. Embeddings for this book have already been created so you can use it out of the box 📦

Book2Vec.ipynb

Use this to see how embeddings for '1984' were made or create your own for another book!

Limitations

This can't yet do summaries or continue a conversaion; each question gets a single answer. Contributions for these are very welcome!

Potential improvements

Save good outputs to use as examples in the prompt (few shot).

How can I talk to another book?

• Create a vector store with embeddings using Book2Vec.ipynb
• Upload to https://huggingface.co as a dataset (recommended)
• Use Talk2Book.ipynb with the vector store you created

And make it an app:

• Duplicate this Hugging Face Space (or just copy the code in app.py) and change the vector store to the one you created in step 2

Contributing

• Fork
• Install requirements: pip install -r requirements.txt (also at the top of each notebook)
• Install nbdev: pip instal nbdev
• Make changes, run your notebooks

In the terminal, before each commit:

• Run nbdevinstall_hooks to clean the notebooks (removes metadata)

And finally:

• Submit your PR

Using Codespaces/VSCode

Everything you need will be installed when you open Codespaces/VSCode; specified in .devcontainer/

Notes for Codespaces:

• Currently Jupyter notebook doesn’t work on Codespaces for an unknown reason, or at least I can’t, so you’ll have to use JupyterLab
• Open with jupyter lab --NotebookApp.allow_origin='*' --NotebookApp.ip='0.0.0.0'
• For more info on using see https://code.visualstudio.com/docs/datascience/notebooks-web