Roman Lutz

Results 458 comments of Roman Lutz

Do you expect a different schema per message sent, or just one schema for one target? I think that's the main decision point. If it's the latter then we can...

Re: Response target not working with the scorer We don't currently enforce a json format, we just use the other option (I don't recall the name offhand, but switching to...

> The specific bug is that the Responses API doesn't use `response_format` in the body, but the alternative I show. There are also two options: `json_object` which doesn't enforce a...

> Since we are taking a private object and making it public, we should review its API to make sure we like. Adding to the list: tests!

I know some people felt strongly about this at some point 🤣 Personally, I'm happy with it as is. The naming improvement is truly a small one if at all....

Indeed! Have you come across options?

Given the difficulty with determining which one the objective score actually is, I am leaning towards just displaying all scores and getting rid of the flag as it's not actually...

Currently, I would expect the difference between `include_auxiliary_scores` being True and False to be - True: all scores are printed - False: only the objective scores are printed That's not...

From what I understand nanoGCG is for causal models or am I misremembering? Nevermind, I just came to that conclusion from a cursory glance a few days ago because of...

Please do! Nobody is on it to the best of my knowledge.