No, it isn't. It generates 1 token at a time in a loop until the response is finished. It's a highly serialized task. Parallelization increases the throughput of how many queries can be processed simultaneously, but you wouldn't be able to speed up a single query.