---
dataset_info:
  features:
  - name: prompt_text
    dtype: string
  - name: prompt_ids
    sequence: int64
  - name: decode_ids
    sequence: int64
  - name: prompt_pattern
    sequence:
      sequence: int64
  - name: decode_pattern
    sequence:
      sequence: int64
  splits:
  - name: train
    num_bytes: 483379455
    num_examples: 32808
  download_size: 46650593
  dataset_size: 483379455
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

- prompt_ids: `[ 1593, 10, 3547, 6180, 7836, 489, 33, 66, 1 ]`	
- prompt_pattern: `[ [ 14, 12, 17, 17, 17, 17, 10, 4, 12 ], [ 15, 17, 18, 18, 18, 5, 21, 12, 15 ], [ 2, 7, 22, 23, 23, 23, 31, 2, 31 ], [ 8, 8, 2, 2, 2, 2, 28, 7, 8 ], [ 0, 12, 7, 13, 13, 7, 11, 8, 11 ], [ 2, 19, 14, 8, 14, 14, 13, 16, 4 ] ]`


[ 14, 12, 17, 17, 17, 17, 10, 4, 12 ]表示 9 个 prompt ids 在encoder第一个moe 层激活的 expert 索引值。例如 token 1593 在encoder 第一层moe 激活的 expert 是 14.