alishanamin/Burushaski-translator-model

import os

import torch from transformers import MarianMTModel, MarianTokenizer

Load your model

print("Loading model...") model_path = os.path.dirname(os.path.abspath(file)) tokenizer = MarianTokenizer.from_pretrained(model_path) model = MarianMTModel.from_pretrained(model_path) device = "cuda" if torch.cuda.is_available() else "cpu" model = model.to(device) print(f"✅ Model loaded! Using: {device}")

Translation function

def translate(text): inputs = tokenizer( text, return_tensors="pt", padding=True, truncation=True, max_length=128 ).to(device) with torch.no_grad(): output = model.generate( **inputs, max_length=128, num_beams=4, early_stopping=True ) return tokenizer.decode(output[0], skip_special_tokens=True)

Test translations

print("\nTest translations:") print("-" * 50) test_sentences = [ "my name is nouman", ]

for sentence in test_sentences: translation = translate(sentence) print(f"EN: {sentence}") print(f"BUR: {translation}") print()