Can one optimize a prompt that times out? My Wikipedia enrichment ER test of 150 records started timing out and I tried setting three minute timeouts but it just won’t finish.

go to 8 minutes. Gemini Pro is pretty slow. Do you have tool calls as well?

I don't have tool calls, no Gemini just works its magic. They are a possibility in the future to retrieve graph context - like the officers or employees of companies.

also they are sunsetting 2.0 family soon. so maybe time to move up and see if it solves

Oh my god

I presume so? It seems to be a pretty powerful model with a great agent harness.

2.5 flash is really good and 3.0 family is sota.

2.5 flash lite drops integer tracking like 15% compared with more like 1% of 2.0 flash lite. It’s a serious issue for me. I need to prompt optimize.

integer tracking? why would you use lite models? weren't you using 2.0 PRO?

I pay just $100/month for development, 2.0 Flash is just as good as 2.5 Pro. Did its price come down?

same. But cheaper than pro models

You have to use integer IDs with LLMs. UUIDs are trash. I build graphs in prompts and do entity resolution.

why not alphanumeric?

Don’t work as well. Just a fact. BAML documented this recently in the new docs.

sorry i am confused, best of luck though. You will figure it out!

Here’s an indirect contribution to BAML that explains: https://gist.github.com/rjurney/9d47b192d709703fe6f46e9cffb88033