Turing Explorer

Welcome to Turing Explorer! 😄

A mechanistic interpretability tool to understand the internals of Turing-LLM, a large language model.

Please feel free to contact me using my email for any queries or for a chat 🙂: [email protected]

Live on: https://turingexplorer.com

Local Use

First, download the files for this app by doing the following:

• Click the <> Code button in this repository.
• Then on the dropdown, click Download ZIP.
• Once downloaded, unzip the file (probably named Turing-LLM-Explorer-main.zip).

To use the interface (front-end):

• Navigate to front-end/ in your terminal and type npm install to download the front-end's dependencies.
• Once installed, type npm run start to run the app locally.

To run the logic (back-end):

• Download the following files and place them in their corresponding target locations:
  • Turing-LLM-1.0-254M can be downloaded here: https://www.kaggle.com/models/danieljamesdavies/turing-llm-1.0-254m. Once downloaded, place model_1722550239_03986.pt in back-end/TuringLLM/.
  • Sparse Autoencoders for All Turing-LLM Layers can be downloaded here: https://www.kaggle.com/datasets/danieljamesdavies/turing-llm-sparse-autoencoders. Once downloaded, place the folder sae/ inside back-end/SAE/.
  • Turing-LLM Explorer Latent Data can be downloaded here: https://www.kaggle.com/datasets/danieljamesdavies/turing-llm-explorer-latent-data. Once downloaded, place the folder latent_data in /back-end. Overwrite the original latent_data file if requested.
• Navigate to back-end/ in your terminal and type pip install -r requirements.txt to download the back-end's dependencies.
• Once installed, type python app.py (or python3 app.py) to run the app locally