@staticmethod
    def verify_greedy_one_batch(verify_guess_tokens, next_guess_tokens):
        gg = 0
        for eg, guess_tokens in enumerate(verify_guess_tokens):
            correct = next_guess_tokens[eg]
            guess = guess_tokens
            if guess != correct:
                break
            gg += 1

from ...utils.log.logging import logger  # 新增
class PluginManager:
    def __init__(...):
        self.all_token_num = 0
        self.all_decode_count = 0
        self.all_prefill_count = 0
        ...
    @timer.track_time_async('generate_token')
    def generate_token(self, input_metadata: InputMetadata):
        ...
        span_end(prof)

        if not input_metadata.is_dummy_batch:
            if not input_metadata.is_prefill:
                for i in range(input_metadata.batch_size):
                    next_tokens = generation_output.token_ids[i]
                    if -1 in next_tokens:
                        first_neg_one_index = np.argmax(next_tokens == -1)
                        next_tokens = next_tokens[:first_neg_one_index]
                    self.all_token_num += len(next_tokens)
                    self.all_decode_count += 1
            logger.error(f"self.all_token_num is {self.all_token_num}, self.all_decode_count is {self.all_decode_count}, self.all_prefill_count is {self.all_prefill_count}. Ratio is {self.all_token_num / self.all_decode_count}")

            if input_metadata.is_prefill:
                for _ in range(input_metadata.batch_size):
                    self.all_prefill_count += 1
                    
        generation_output.trace_ids = trace_ids