Parst5 Base

🚀 Monolingual Persian T5 Model

A monolingual T5 model for Persian, trained on the OSCAR 21.09 (https://oscar-corpus.com/) corpus using a self-supervised method. The model was pre-trained on a 35 Gig deduplicated version of Persian data.

🚀 Quick Start

This model is similar to the English T5 model but is specifically designed for Persian. You may need to fine-tune it for your specific task.

💻 Usage Examples

Basic Usage

from transformers import T5ForConditionalGeneration,AutoTokenizer

import torch

model_name = "Ahmad/parsT5-base"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)

input_ids = tokenizer.encode('دانش آموزان به <extra_id_0> میروند و <extra_id_1> میخوانند.', return_tensors='pt')
with torch.no_grad():
    hypotheses = model.generate(input_ids)
for h in hypotheses:
    print(tokenizer.decode(h))

📚 Documentation

Training Details

• Steps: 725000
• Accuracy: 0.66

Training More

To train the model further, please refer to its GitHub repository at: https://github.com/puraminy/parsT5