Simon Willison’s Weblog

Subscribe

Thursday, 25th July 2024

wat (via) This is a really neat Python debugging utility. Install with pip install wat-inspector and then inspect any Python object like this:

from wat import wat
wat / myvariable

The wat / x syntax is a shortcut for wat(x) that's quicker to type.

The tool dumps out all sorts of useful introspection about the variable, value, class or package that you pass to it.

There are several variants: wat.all / x gives you all of them, or you can chain several together like wat.dunder.code / x.

The documentation also provides a slightly intimidating copy-paste version of the tool which uses exec(), zlib and base64 to help you paste the full implementation directly into any Python interactive session without needing to install it first.

# 6:58 pm / python

Button Stealer (via) Really fun Chrome extension by Anatoly Zenkov: it scans every web page you visit for things that look like buttons and stashes a copy of them, then provides a page where you can see all of the buttons you have collected. Here's Anatoly's collection, and here are a few that I've picked up trying it out myself:

Screenshot showing some buttons I have collected, each with their visual appearance maintained

The extension source code is on GitHub. It identifies potential buttons by looping through every <a> and <button> element and applying some heuristics like checking the width/height ratio, then clones a subset of the CSS from window.getComputedStyle() and stores that in the style= attribute.

# 7:40 pm / chrome, css, extensions, javascript

AI crawlers need to be more respectful (via) Eric Holscher:

At Read the Docs, we host documentation for many projects and are generally bot friendly, but the behavior of AI crawlers is currently causing us problems. We have noticed AI crawlers aggressively pulling content, seemingly without basic checks against abuse.

One crawler downloaded 73 TB of zipped HTML files just in Month, racking up $5,000 in bandwidth charges!

# 8:02 pm / eric-holscher, ethics, ai, read-the-docs

Introducing sqlite-lembed: A SQLite extension for generating text embeddings locally (via) Alex Garcia's latest SQLite extension is a C wrapper around the llama.cpp that exposes just its embedding support, allowing you to register a GGUF file containing an embedding model:

INSERT INTO temp.lembed_models(name, model)
  select 'all-MiniLM-L6-v2',
  lembed_model_from_file('all-MiniLM-L6-v2.e4ce9877.q8_0.gguf');

And then use it to calculate embeddings as part of a SQL query:

select lembed(
  'all-MiniLM-L6-v2',
  'The United States Postal Service is an independent agency...'
); -- X'A402...09C3' (1536 bytes)

all-MiniLM-L6-v2.e4ce9877.q8_0.gguf here is a 24MB file, so this should run quite happily even on machines without much available RAM.

What if you don't want to run the models locally at all? Alex has another new extension for that, described in Introducing sqlite-rembed: A SQLite extension for generating text embeddings from remote APIs. The rembed is for remote embeddings, and this extension uses Rust to call multiple remotely-hosted embeddings APIs, registered like this:

INSERT INTO temp.rembed_clients(name, options)
  VALUES ('text-embedding-3-small', 'openai');
select rembed(
  'text-embedding-3-small',
  'The United States Postal Service is an independent agency...'
); -- X'A452...01FC', Blob<6144 bytes>

Here's the Rust code that implements Rust wrapper functions for HTTP JSON APIs from OpenAI, Nomic, Cohere, Jina, Mixedbread and localhost servers provided by Ollama and Llamafile.

Both of these extensions are designed to complement Alex's sqlite-vec extension, which is nearing a first stable release.

# 8:30 pm / c, sqlite, rust, alex-garcia, embeddings

Our estimate of OpenAI’s $4 billion in inference costs comes from a person with knowledge of the cluster of servers OpenAI rents from Microsoft. That cluster has the equivalent of 350,000 Nvidia A100 chips, this person said. About 290,000 of those chips, or more than 80% of the cluster, were powering ChartGPT, this person said.

Amir Efrati and Aaron Holmes

# 9:35 pm / generative-ai, openai, chatgpt, ai, llms