---
library_name: transformers
datasets:
- ai4bharat/samanantar
language:
- en
- te
metrics:
- bleu
base_model:
- aryaumesh/english-to-telugu
---



---

# English-to-Telugu Translation Model  

## Overview  
This project is a deep learning-based English-to-Telugu translation model trained on a custom dataset. It uses Hugging Face Transformers for NLP and was developed in Google Colab. The model can be used for translating sentences with improved contextual accuracy.  

## Features  
✅ Translates English text to Telugu  
✅ Trained on a custom bilingual dataset  
✅ Uses Transformer-based model
✅ Implemented and trained in Google Colab  
✅ Can be fine-tuned for better accuracy  

## Tech Stack  
- **Programming Language**: Python  
- **Framework**: Hugging Face Transformers  
- **Model**: mBART (Fine-tuned)  
- **Libraries**:  
  - transformers (Hugging Face)  
  - `torch` (PyTorch)  
  - `sentencepiece` (Tokenization)  
- **Platform**: Google Colab  

## Dataset  
- Used a custom English-Telugu parallel corpus 
- Preprocessed using:  
  - **Tokenization** (SentencePiece / WordPiece)  
  - **Lowercasing & Cleaning**  
  - **Removing noisy data**  

## Model Training  
Training was done in Google Colab using a GPU. Here’s a snippet of the fine-tuning process:  

from transformers import MarianMTModel, MarianTokenizer, Trainer, TrainingArguments

# Load pre-trained model & tokenizer
model_name = "aryaumesh/english-to-telugu"  # Base model
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)

# Preprocess dataset (example)
def encode_data(texts):
    return tokenizer(texts, padding=True, truncation=True, return_tensors="pt")

# Training arguments
training_args = TrainingArguments(
    output_dir="./results",
    per_device_train_batch_size=8,
    num_train_epochs=3,
    save_steps=1000,
    save_total_limit=2,
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=custom_dataset,
)

trainer.train()  

## Run the Model  
  
def translate(text):
    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
    translated = model.generate(**inputs)
    return tokenizer.decode(translated[0], skip_special_tokens=True)

english_text = "Good morning, how are you?"
telugu_translation = translate(english_text)
print("Translated Text:", telugu_translation)
   
## Future Improvements  
🔹 Train on a larger dataset for better accuracy  
🔹 Optimize inference speed for real-time use  
🔹 Deploy as a cloud-based API (AWS/GCP)  

---