crf_decoding 层 transition 矩阵维度问题

您好，我在使用默认 LAC 模型分词时，把模型打出来，发现 viterbi decoding 的 transition 矩阵维度不同，这里不应该是一个 num_tags*num_tags 的矩阵吗？多出来的 2 是因为什么？文档哪里有解释吗？

Start with the crf_decoding layer and the default LAC model's Viterbi decoding output shown in the issue. Trace how the transition matrix is created and compare its dimensions with num_tags*num_tags. Done means explaining the extra 2 and adding or pointing to documentation that resolves the discrepancy.

Written by the indexing model from the issue text.