neuronpedia/python-code-simplified

---
license: mit
---

## What This Is
This is a dataset of "simplified" Python Code from https://huggingface.co/datasets/iamtarun/python_code_instructions_18k_alpaca.
Our simplification attempts to remove all comments, and reduce these names/strings to a single letter without impacting its structure/logic.
You should use the "minimized" field. The "original" field is the original output from `iamtarun/python_code_instructions_18k_alpaca`.

## Why It Exists
We used this dataset to generate dashboards for OpenAI's [circuit_sparsity](https://huggingface.co/openai/circuit-sparsity) model, which was trained on unspecified Python Code.
It was necessary to create a "simplified" version of Python code samples, because the original Python code has many variable names, comments, and strings that the model likely wasn't trained on, resulting in dashboards that are not very interpretable.

## Where It's Used
Example dashboard on Neuronpedia here: https://www.neuronpedia.org/circuitgpt-python/6-mlp/5543