metadata
library_name: transformers
tags:
- Fill Mask
- Persian
- BERT
Model Details
Model Description
This model is fine-tuned for the task of masked language modeling in Persian. The model can predict missing words in Persian sentences when a word is replaced by the [MASK] token. It is useful for a range of NLP applications, including text completion, question answering, and contextual understanding of Persian texts.
- Developed by: Behpouyan
- Model type: Encoder
- Language(s) (NLP): Persian
How to Get Started with the Model
from transformers import AutoTokenizer, AutoModelForMaskedLM
import torch
# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("Behpouyan/Behpouyan-Fill-Mask")
model = AutoModelForMaskedLM.from_pretrained("Behpouyan/Behpouyan-Fill-Mask")
# List of 5 Persian sentences with a masked word (replacing a word with [MASK])
sentences = [
"این کتاب بسیار <mask> است.", # The book is very <mask
"مشتری همیشه از <mask> شما راضی است.", # The customer is always satisfied with your <mask
"من به دنبال <mask> هستم.", # I am looking for <mask
"این پروژه نیاز به <mask> دارد.", # This project needs <mask
"تیم ما برای انجام کارها <mask> است." # Our team is <mask to do the tasks
]
# Function to predict masked words
def predict_masked_word(sentence):
# Tokenize the input sentence
inputs = tokenizer(sentence, return_tensors="pt")
# Forward pass to get logits
with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits
# Get the position of the [MASK] token
mask_token_index = torch.where(inputs.input_ids == tokenizer.mask_token_id)[1].item()
# Get the predicted token
predicted_token_id = torch.argmax(logits[0, mask_token_index]).item()
predicted_word = tokenizer.decode([predicted_token_id])
return predicted_word
# Test the model on the sentences
for sentence in sentences:
predicted_word = predict_masked_word(sentence)
print(f"Sentence: {sentence}")
print(f"Predicted word: {predicted_word}")
print("-" * 50)