search

X-PLUG / mPLUG-DocOwl

mPLUG-DocOwl: Modularized Multimodal Large Language Model for Document Understanding
Apache License 2.0
1.51k stars 99 forks source link

Spelling errors in DocStruct4M, 'multi_grained_text_localization.jsonl' #34

Closed weichenma15 closed 7 months ago

weichenma15 commented 7 months ago

All the question prompts are extracted from DocStruct4M, 'multi_grained_text_localization.jsonl' as below,

[
  "Give the bounding box of the text",
  "Predict the bounding box of the text",
  "Detect the text in the bounding box",
  "Identify the text within the bounding box",
  "Recognize the text in the bounding box",
  "Locate the postion of the text"
]

In the last column, 'postion' should be replaced with 'position'. I wonder whether it matters for training the MLLM, because the error amount is significantly high.

HAWLYQ commented 7 months ago

Hi, @chrysanthemum-515 Thanks a lot for your careful check and kind reminder. I think this minor error will not significantly influence the model due to the strong language understanding ability of LLM. If you feel risky, you can revise this spelling error for model training and evaluation.

weichenma15 commented 7 months ago

Hi, @chrysanthemum-515 Thanks a lot for your careful check and kind reminder. I think this minor error will not significantly influence the model due to the strong language understanding ability of LLM. If you feel risky, you can revise this spelling error for model training and evaluation.

Thanks for reply.