Views
No views yet
| Translation Method | Accuracy |
|---|---|
| Direct Bangla (no translation) | 35.3% |
| BanglaT5-Base | 59.8% |
| Google Translate | 76.5% |
| BanglaSTEM-T5 (Ours) | 82.5% |
| Translation Method | Success Rate |
|---|---|
| Direct Bangla (no translation) | 31.0% |
| BanglaT5-Base | 59.0% |
| Google Translate | 72.0% |
| BanglaSTEM-T5 (Ours) | 79.0% |
pip install transformers torch1from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
2
3# Load model and tokenizer
4tokenizer = AutoTokenizer.from_pretrained("reyazul/BanglaSTEM-T5")
5model = AutoModelForSeq2SeqLM.from_pretrained("reyazul/BanglaSTEM-T5")
6
7# Translate Bangla to English
8bangla_text = "একটি পাইথন ফাংশন লিখুন যা একটি তালিকার সর্বোচ্চ মান খুঁজে বের করে।"
9inputs = tokenizer(bangla_text, return_tensors="pt", padding=True)
10outputs = model.generate(**inputs, max_length=128, num_beams=4)
11english_translation = tokenizer.decode(outputs[0], skip_special_tokens=True)
12
13print(english_translation)
14# Output: "Write a Python function that finds the maximum value in a list."1# For more accurate translations
2outputs = model.generate(
3 **inputs,
4 max_length=256,
5 num_beams=5,
6 early_stopping=True,
7 temperature=0.7,
8 do_sample=False
9)1@inproceedings{hasan2026banglastem,
2 title={BanglaSTEM: A Parallel Corpus and Term-Weighted Evaluation for Technical Bangla-English Translation},
3 author={Hasan, Kazi Reyazul and Al Islam, ABM Alim and Adnan, Muhammad Abdullah},
4 booktitle={Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026)},
5 pages={403--412},
6 year={2026}
7}
8