This model is trained with Khmer annotation dataset in kaggle (https://www.kaggle.com/datasets/keatchakravuth/khmer-annotation). This data set contains around <b>26K Khmer sentences</b>. The images from this dataset are also used to build OCR model by using this model as the decoder as well.
This is a decoding model using Roberta. <b>Using this model as a decoder with Vision Encoder Decoder Model (ViED) is recommended</b>. This model fully supports for Khmer language with Byte-Pair Encoding (BPE) for character-level. Please refer to this link (https://huggingface.co/channudam/khmer-xlm-roberta-base) for Byte-Pair Encoding (BPE) for word-level.
