MaiBERT_TF: Mithila Language Pretrained Transformer (TensorFlow)

Overview

Model Details

• Model Name: MaiBERT_TF
• Architecture: Transformer
• Pretrained Framework: TensorFlow
• Language: Mithila
• Vocabulary Size: [VOCAB_SIZE]
• Model Size: [MODEL_SIZE]
• Tokenizer Used: [TOKENIZER_NAME]

How to Use

1. Installation: Install the required libraries using the following command:
   pip install transformers tensorflow
2. Loading the Model: Load the pretrained model using the Hugging Face Transformers library:
   python

   1from transformers import TFAutoModelForMaskedLM, AutoTokenizer 2 3model = TFAutoModelForMaskedLM.from_pretrained("rockerritesh/maiBERT_TF") 4tokenizer = AutoTokenizer.from_pretrained("rockerritesh/maiBERT_TF")
3. Usage: Once the model and tokenizer are loaded, you can use them for various tasks like text generation, text completion, and more. Here's an example of generating text:
   python

   1input_text = "जानकी मन्दिर पवित्र हिन्दू मन्दिर छी" 2inputs = tokenizer(input_text, return_tensors="tf", padding=True, truncation=True) 3 4output = model.generate(inputs["input_ids"]) 5generated_text = tokenizer.decode(output[0], skip_special_tokens=True) 6print("Generated Text:", generated_text)
4. Fine-Tuning: If you want to fine-tune the pretrained model for a specific task, you can do so by loading the model using TFAutoModelForSequenceClassification and providing your own dataset.

Model Training Details

• Training Data: [TRAINING_DATA_SOURCE]
• Training Objective: Masked Language Modeling (MLM)
• Training Steps: [TRAINING_STEPS]
• Batch Size: [BATCH_SIZE]
• Learning Rate: [LEARNING_RATE]

Citation

@misc{yadav2025maibertspeakmaithili,
      title={Can maiBERT Speak for Maithili?}, 
      author={Sumit Yadav and Raju Kumar Yadav and Utsav Maskey and Gautam Siddharth Kashyap Md Azizul Hoque and Ganesh Gautam},
      year={2025},
      eprint={2509.15048},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2509.15048}, 
}