[kbumsik]

Right, I wouldn't use full table detection to VLM model because they tend to mistake with numbers in table...