Agent Lightning: Train agents with RL (no code changes needed) (github.com)

> Turn your agent into an optimizable beast with ZERO CODE CHANGE (*almost*)!

OP didn’t think to include this very important fine print. Thanks OP!

What actually is this?

All these agent documentations seem to compete for the most complex set of flow charts imaginable without ever mentioning what the Rube Goldberg machine is supposed to accomplish. Given that the real output in open source of these contraptions is zero, it seems that the flow charts are the goal. Some kind of modern art.

do you have benchmarks on tasks with sparse rewards or partial observability? i feel like thats where most "train any agent" claims tend to break down

Parsing entireties of the I/O agent release version, which is the precommit as text prior to evaluation.

Let’s see…excessive emojis and wacky punctuation hmm maybe this whole readme is LLM generated.

So it’s some brittle crap built on verl, which is already pretty much train by config (and makes breaking changes with _every single commit_), with no documentation, no examples, and no clear purpose? Heck yeah Microsoft

Interesting that the project is re-alive on HackerNews! I've posted about the project weeks ago on HackerNews. Got no attention at all haha!

https://news.ycombinator.com/item?id=44861765