I desperately want to be able to get a concrete amount of tokens for my prompt before making a call - things like this make it very hard to request the right amount of max_tokens from longer prompt/generation pairs.