When I started looking into agent swarm observability, I really wanted some real data to work with. I wasn't really able to find much, so I created my own and am releasing it on the @huggingface Hub for others to use. I recorded 19 runs of 18 agents that were all working on building a small Lisp interpreter. I used a custom recorder (more about this at a later date perhaps) which records from underneath the agents rather than just relying on their own reports. So I tracked all the programmes they ran, every file they read or wrote, every model call, every message between them and so on. In some runs I planted faults on purpose (bad commands, rewritten logs, incomplete specs). All in all it's 8.9 million events. 10 runs come with answer keys. 9 are held back as a test set. AFAIK, most multi-agent datasets are ~5 agents large only, and none of them record file reads and writes. So I hope this dataset can help people working on swarm observability and those building tools to help debug multi-agent swarms. Enjoy! (I'll also try to write this up with a bit more detail on my blog later) huggingface.co/datasets/zenm…

4 real production LLM use cases (curated by yours truly). 2x per week. Under 3 minutes to read. Next email going out on Tuesday. Subscribe here: zenml.io/llmops-database

Finally got round to adding in a way to subscribe to my blog / site via email. (Always supported RSS, but seems that's gone out of fashion a bit 🤷 ) Also added in a way to browse some of the blog series I've written over the years instead of just scrolling through a huge long list of posts. Please do subscribe if you want to get an email whenever I write new posts! alexstrick.com/subscribe.htm… alexstrick.com/series.html