loodos/albert-base-turkish-uncased

Turkish ALBERT-Base (uncased) model has 12 repeated encoder layers with 768 hidden layer size trained on uncased Turkish dataset . Currently, Huggingface's tokenizers (which were written in Python) have a bug concerning letters "ı, i, I, İ" and non-ASCII Turkish specific letters . Until it is fixed, we provide a simple text normalization module (TextNormalization() in the code snippet above) to train your model with uncased data . Many thanks to TFRC Team for providing us cloud TPUs on Tensorflow Research Cloud to train our models . The model was created with NFC/NFKC normalization but tokenizer,