A lot of times, models (including claude 3.7) will just stop in the middle of generation because of the output token limit of 4000. Please increase this limit, to at least 8192 or 16384. It would be way better that way.