README.md

metadata

language:
  - en
dataset_info:
  features:
    - name: prompt
      dtype: string
    - name: answer
      dtype: string
  splits:
    - name: spell
      num_bytes: 465046
      num_examples: 1000
    - name: spell_inverse
      num_bytes: 446046
      num_examples: 1000
    - name: contains_char
      num_bytes: 424176
      num_examples: 1000
    - name: contains_word
      num_bytes: 530494
      num_examples: 1000
    - name: orth
      num_bytes: 607118
      num_examples: 1000
    - name: sem
      num_bytes: 608098
      num_examples: 1000
    - name: ins_char
      num_bytes: 560474
      num_examples: 1000
    - name: ins_word
      num_bytes: 775597
      num_examples: 1000
    - name: del_char
      num_bytes: 513247
      num_examples: 1000
    - name: del_word
      num_bytes: 689114
      num_examples: 1000
    - name: sub_char
      num_bytes: 532364
      num_examples: 1000
    - name: sub_word
      num_bytes: 743529
      num_examples: 1000
    - name: swap_char
      num_bytes: 470394
      num_examples: 1000
    - name: swap_word
      num_bytes: 675168
      num_examples: 1000
  download_size: 962103
  dataset_size: 8040865
configs:
  - config_name: default
    data_files:
      - split: spell
        path: data/spell-*
      - split: spell_inverse
        path: data/spell_inverse-*
      - split: contains_char
        path: data/contains_char-*
      - split: contains_word
        path: data/contains_word-*
      - split: orth
        path: data/orth-*
      - split: sem
        path: data/sem-*
      - split: ins_char
        path: data/ins_char-*
      - split: ins_word
        path: data/ins_word-*
      - split: del_char
        path: data/del_char-*
      - split: del_word
        path: data/del_word-*
      - split: sub_char
        path: data/sub_char-*
      - split: sub_word
        path: data/sub_word-*
      - split: swap_char
        path: data/swap_char-*
      - split: swap_word
        path: data/swap_word-*

CUTE

Here is the CUTE benchmark, a benchmark designed for testing LLM's ability to understand the characters within their tokens.

To use this dataset as we did, make sure to use tokenizer.apply_chat_template on the prompt, and then add Answer: " afterwards to the resulting string. For smaller LLMs, you may need to do some post-processing on the final answer.

You can also check out our implementation at: https://github.com/Leukas/cute