Understand Large Language Models

• Visualization of Logistic Lens for bi-directional and causal attention)

• Visualization of attention and predication

• Live demonstration visualizing attention in an transformer

  • Uses ONYX transformer running in your browser

  • Visualizes both next word prediction and the contribution each previous word to final prediction

  • Github repository

Teaching a Large Language Model to Use Tools

• One of the most exciting directions work in LLM is giving them the ability to use computational tools, such as calculators, scientific databases, and real-time web search.

• There are several fully-featured frameworks for doing this, such as LangChain.

• For instructional use, however, it is useful to provide students with a very small and simple code base that they can extend with new tools.

• Thus my tiny framework, gpt_with_tools .

Mazes: Breadth-first, Depth-first, Best-first, and A* Search Demo

• Download AMazer.jar

• Run java -jar AMazer.jar

• Download source zip file AMazer.zip

N-Queens: Randomized Backtracking and Local Search Demo

• Click here to run the Typescript program.

• Download source from Gitlab.

Sokoban Solution Visualizer

• This is not a Sokoban solver, but rather a visualizer for Sokoban problems and solutions.

• See the Help page of the program for details of file formats, options, and downloads.

• Click here to run the Typescript program.

Walksat

• Satisifiability testing by local search.

• Go to the Walksat home page.

Schema

• Schema is a language for concisely stating Boolean logic formulas by using a first-order lisp-based syntax.

• Requires Common Lisp.

• Go to the Schema home page.