- Use Node.JS 22 (I recommend installation via nvm) and pnpm (after Node.JS is installed, using
npm i -g pnpm) - Run
pnpm install - Install ollama from https://github.com/trustedtomato/ollama (until ollama/ollama#2404 gets merged)
- See this page for installation guidance: https://github.com/trustedtomato/ollama/blob/main/docs/development.md
- For developing GBNF in Visual Studio Code, install https://marketplace.visualstudio.com/items?itemName=iddar.gbnf-highlighter
In order to have a good understanding of the codebase, it is recommended to start with the entry point of the test suite runner: src/tests/index.ts. From there, you can explore the rest of the codebase by following the imports.
Some of the more interesting files include:
- The test suite runner: src/tests/index.ts
- The ROS node: src/rosnode.ts
- One of the simpler algorithms, for example: src/algorithms/alg-cot-zeroshot.ts
- The utility function called chat, which provides an abstraction over the Ollama API: src/utils/chat.ts
- The base definitions for the output grammars: src/utils/grammars.gbnf
- To check how it is used to create a grammar, see src/algorithms/alg-cot-zeroshot.ts
- Run ollama by going into its directory and running
while true; do ./ollama serve > /dev/null 2>&1 || sleep 1; done- The script makes sure to restart the Ollama server in case of a crash, which does happen sometimes
- The script also suppresses the output of the Ollama server, which could
potentially be very verbose and affect the performance metrics of the test
suite. For development though, you could remove it, in which case you'd
want to remove the
> /dev/null 2>&1part of the command
- Run
npm run test- You can run
DEBUG="app:test,app:test:*" npm run testto see the test suite output - You can run
DEBUG="app:*" npm run testto see the test suite output and the debugging output of all other parts of the codebase - To run the test suite multiple times, you could use
while true; do npm run test; done
- You can run