Hello, world

Welcome to my blog! This is a sample post. Replace it with your own writing, or keep it as a template.

How to write a new post

Create a new file in the _posts folder named YYYY-MM-DD-your-title.md, copy the header block (the lines between the --- markers) from this file, and write your post below it in Markdown. When you push the file to GitHub, the site rebuilds and the post appears on the blog page.

What you can use

You can write bold and italic text, links, lists, and quotes:

Interpretability is the study of what a model has learned, not only what it outputs.

Code blocks are highlighted:

import torch

# mean activation of one attention head over a batch
head_mean = attn[:, layer, head].mean(dim=0)

And math is rendered with MathJax, for example the logit lens projection \(\hat{y}_\ell = W_U \, \mathrm{LN}(h_\ell)\), or as a display equation: