2 ms·
> On a per-call basis, the wrappers are pure python ifs and such, measured in ms easily Ah that's good to know when I first saw this posted yesterday I was wo
by anentropic 5mo ago
> On a per-call basis, the wrappers are pure python ifs and such, measured in ms easily
Ah that's good to know
when I first saw this posted yesterday I was wondering that, kind of assumed maybe it was doing extra LLM calls to make judgements
- zambelli 5mo agoRetry nudges do generate an extra LLM call, and those average extra calls time impacts are captured in the eval data. But that's the difference between the call failing and succeeding (eventually). On successful calls the presence of forge should be unnoticeable.