Mongolian cyrillic text classification with modern tensorflow and some fine tuning on TugsTugi's BERT model.
On TPU mode, loading checkpoints from the file system doesn't supported by the bert and bucket should be used.
Fine tuning mongolian BERT on TPU, You need own bucket in order to finetune on TPU
Fine tune a mongolian BERT on GPU, a lot of computation needed, a low batch size matters due to memory limit
No 02, Pretrained Word2Vec initialization from Facebook's fasttext, kind of transfer learningish. Embedding layer is not trainable in this case and with trainable embedding layer
No 03, 1D Convolution and multiple 1D convnets
Visualize RNN neuron firing in text generation
No 05, LSTM with Attention, visualization of attention scores in text classification
No 06, Classification with Mongolian BERT and Tensorflow 2.0, with frozen bert layers
No 07, Classification with Mongolian BERT large and HuggingFace and Tensorflow 2
Sequence loss in keras and tf2
Variational Auto Encoder for Mongolian text
Predict next word, greedy text generation
word2vec initialization, 1D Convolution, RNN variants, Attention, Some weights visualization for reasoning, Transformer, Techniques to handle longer texts and so on...
- Mongolian BERT models https://github.com/tugstugi/mongolian-bert
- Mongolian NLP https://github.com/tugstugi/mongolian-nlp
- Eduge classification baseline using SVM https://colab.research.google.com/github/tugstugi/mongolian-nlp/blob/master/misc/Eduge_SVM.ipynb
- News crawler https://github.com/codelucas/newspaper