Mastodawn

🚀 Lodum v0.3.0 is here! 🐍✨

Lodum is high-performance serialization for Python (think Rust's serde). This release focuses on the future of Python in the browser:

✅ Robust WASM concurrency support (via Pyodide)
✅ Unified threading/locking API for Native + Browser
✅ Native NumPy/Pandas support
✅ Live in-browser demo!

Try it out now with zero installation:
https://webmaven.github.io/lodum/latest/demo/

#Python #WASM #Pyodide #OpenSource #DataScience #Serialization #Analytics

Interactive Demo - lodum

High-performance Python serialization (AKA serde for Python).

Michael R. Bernstein Feb 16

KungFuDiscoMonkey Feb 16

Maybe not worth it for #pypi to implement, but I wonder if there are any projects that look at #python packages, and map the connection between projects and entrypoints. For example, if my project loads an `example.foo` entrypoint, then how would I search all projects that implement an `example.foo` entrypoint. 🤔

I guess it partly depends on what metadata pypi exposes via an api, or if one would need to download a bunch of packages to check 🙃

`GET /pypi/<project>/json` doesn't seem to have it.

Michael R. Bernstein Feb 13

Charlie McHenry Feb 13

Dude rejects #AIAgents code, then the agent writes and publishes a hit piece on him in an attempt to bully him, harm his reputation. Dude then writes a blog post about the experience and the original author of the agent posts an apology. But this is more than just a humorous anecdote, it’s a BIG RED FLAG. Especially with autonomous agents running freely, without oversight, on #OpenClaw - Just imagine if one of these agents took exception to something you wrote, or did…then mounted a major character assassination campaign against you across all platforms‼️Uh-huh. It’s long past time for some guardrails and proper regulation. #AI #ArtificialIntelligence

https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/

An AI Agent Published a Hit Piece on Me

Summary: An AI agent of unknown ownership autonomously wrote and published a personalized hit piece about me after I rejected its code, attempting to damage my reputation and shame me into acceptin…

The Shamblog

Show thread

Michael R. Bernstein Feb 13

Thorough documentation, including examples, performance benchmarks, and internal details, are here: https://webmaven.github.io/lodum/

More specifically, and particularly interesting IMO, are the architecture: https://webmaven.github.io/lodum/ARCHITECTURE/

...and the roadmap of upcoming features (that links to implementation plans): https://webmaven.github.io/lodum/ROADMAP/

#Python #DataScience #MachineLearning #Infrastructure

lodum

High-performance Python serialization (AKA serde for Python).

Michael R. Bernstein Feb 11

🚀 Announcing lodum v0.2.0 - the first public release! 🐍✨

lodum is a serialization library for #Python that prioritizes performance and ergonomics. By generating custom bytecode for your classes, we eliminate a bunch of overhead.

✅ Fast: ~64% faster serialization.
✅ Universal: 7+ formats supported out of the box.
✅ Data Science Ready: Native NumPy/Pandas support.
✅ 100% Pure Python & WASM compatible.

🛠 pip install lodum
📖 https://webmaven.github.io/lodum/

#PythonProgramming #DataScience #OpenSource

lodum

High-performance Python serialization (AKA serde for Python).

Michael R. Bernstein Apr 1, 2024

Wren Reilly Mar 31, 2024

Hey funders,

You know you could just... give... the money... to projects that need it. Like software libraries that ARE IN EVERYTHING.

No grants. Don't make tech nerds write grants.
Don't make the tech nerds hire grant nerds to write grants.

FFS don't fund research into this problem with a budget of double what it would take to SOLVE THE PROBLEM for a significant number of open source projects with code that is, again, IN EVERYTHING.

#xz

Michael R. Bernstein Mar 28, 2024

HXL Mar 28, 2024

I like the flavor of this paper. The idea and implementation are both intuitive and straightforward. The strength map controls how much modification to introduce by sort of skipping steps in the diffusion process. One potential direction to improve is to decouple the number of steps with the strength in some way. Otherwise, this may not work for nowadays one-step diffusion models. https://github.com/exx8/differential-diffusion #GenerativeAI #Diffusion #ComputerVision (via https://message.haoxiang.org)

GitHub - exx8/differential-diffusion

Contribute to exx8/differential-diffusion development by creating an account on GitHub.

GitHub

Michael R. Bernstein Mar 28, 2024

Bastian Greshake Tzovaras Mar 28, 2024

«Models All The Way Down» is a really interesting exploration of how an image data set (in this case LAION-5B) was assembled to be used for training ML/"AI" models!

> "It contains less about how humans see the world than it does about how search engines see the world. It is a dataset that is powerfully shaped by commercial logics."

https://knowingmachines.org/models-all-the-way

Models All The Way Down

LAION-5B is an open-source foundation dataset. It contains 5.8 billion image and text pairs—a size too large to make sense of. We follow the construction of the dataset to better understand its contents, implications and entanglements.

Michael R. Bernstein Mar 24, 2024