Skip to content

v2.0.0
Compare
Choose a tag to compare
@rgbkrk rgbkrk released this 25 Mar 02:25
v2.0.0
91efebf
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

genai @ 2.0.0

Generative AI Tooling for IPythonic Platforms.

Install genai in your notebook environment now!

🦾 Get GPT help with code, SQL queries, DataFrames, Exceptions, and more in IPython.

🌍 Supports all Jupyter environments, including IPython, JupyterLab, Jupyter Notebook, and Noteable.

%pip install genai
%load_ext genai

Enhancements:

Added

  • 🔄 Keep conversations flowing with %%assist (#66)
  • 🖼️ Emit suggestions as Markdown instead of creating new cells (#66)
  • 🚀 Model selection made easy with the --model flag for %%assist (#65)
  • 💡 Introducing GenaiMarkdown – a dynamic Markdown display (#61)
  • 📝 Create a %%prompt magic for setting the default prompts for assistance and exceptions (#71, #69)

Changed

  • 🧪 Craft a more ipythonic context manager (#62, #66)

    • Meet the new Context class: capture IPython history and make it ChatCompletion-friendly
    • Farewell get_historical_context, hello build_context: context construction using the new Context class
    • Reduce messages sent to GPT models by trimming based on estimated number of tokens (#57)

  • 🎯 Type annotations step in! (#59)

Improved

  • 📏 Token length checks now available in %%assist (#57)
  • 🧹 Code refactoring: introducing craft_message, repr_genai_pandas, and repr_genai for more organized and readable code
  • 📈 Enhanced pandas support: optimized DataFrame and Series representation for Large Language Model consumption using Markdown format
  • 💰 Token management: a new module tokens.py featuring num_tokens_from_messages and trim_messages_to_fit_token_limit to help you stay within model limitations and budget
  • 📚 Update assist magic documentation (#70)

Removed

  • 🚫 %%assist no longer generates new code cells. It now creates Markdown output instead (#66)
    • Relatedly, in-place is no longer an option since we do not change the cells

Changes:

  • craft_user_message now relies on the new craft_message function
  • craft_output_message has been upgraded to use the new repr_genai function
  • get_historical_context now sports an additional model parameter and utilizes tokens.trim_messages_to_fit_token_limit
  • For clarity, the ignore_tokens list now uses the term "first line" instead of "start"
  • GPT-4 token counting and message trimming now supported in tokens.py
Assets 2