weird how every time you see this image on twitter it has a ton of retweets

I had no idea Italians held such strong views on open source licensing

RT @shoyer: One lesson for writing NumPy/JAX code that took me a surprisingly long time to learn is to preserve array dimensionality whenev….

Huge new features to be excited about:.- better chart saving via vl-convert.- more convenient method-based chart syntax.- new offset channel for jitter & grouped charts.- more flexible interactions API.- and much more!.

Excited to see the release of Vega-Altair version 5.0!. Thanks so much to all the contributors who made this happen!

The weirdest thing about having solar panels on your house is when you realize that a cloudy day costs you like $5.00.

four published papers stapled together.

There are four main modes of remote work: working from home, working from “home”, “working” from home, and “working” from “home”.

RT @vegafusion_io: 🎉 VegaFusion 1.2 is out! This release introduces support for saving Altair charts to HTML, JSON, and static image files.….

A recent project I've been working on: ml_dtypes is a Python package containing a stand-alone implementation of bfloat16 and other NumPy dtypes used in AI & ML — the package is now used by JAX, Tensorflow, OpenXLA, and others!

A Magritte-inspired art installation at the BART station this morning

Jake, 1988: *plays with transformers all day*.Jake, 2023: *plays with transformers all day*. How far I’ve come….

Made it out this weekend for a winter camping excursion at Lake Alpine! Dropped to -10C outside at night, but we ate well and slept warm in our snow cave

Took a while but I think I got it

sister-in-law stopped by this morning asking if I knew how to make some bar charts from some survey data.

I knew this guy once who was a gym teacher; every day he'd get home from work and be like "I can't wait to get out of these work clothes". Then he'd change from his sweats into a bathrobe.

I’ve been walking past this sign in a neighbor’s yard every day for weeks, and I still have no idea who’s supposed to be reading it

For runnable versions of all the code snippets in this thread, see jax_vmap.ipynb in this gist:

And this is what jax.vmap gives you: a way to automatically create efficient batched versions of your functions – that will lower to fast vectorized computations – without having to re-write your code by hand. You can read more about vmap in the JAX docs:

The details differ slightly — for example, xla_call comes from the fact that einsum is jit compiled — but the essential steps in the computation match more-or-less exactly: dot_general(), then add(), then tanh(), then reduce_sum().