Instructions to use helloboyn/imd-tri with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use helloboyn/imd-tri with Transformers:
# Use a pipeline as a high-level helper # Warning: Pipeline type "translation" is no longer supported in transformers v5. # You must load the model directly (see below) or downgrade to v4.x with: # pip install "transformers<5.0.0" from transformers import pipeline pipe = pipeline("translation", model="helloboyn/imd-tri", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForSeq2SeqLM model = AutoModelForSeq2SeqLM.from_pretrained("helloboyn/imd-tri", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
IMD Weather Alert MT
English to Hindi, Maithili and Urdu translation built for weather alerts. Turn India Meteorological Department (IMD) style bulletins and warnings into clear Hindi, Maithili and Urdu, with every number and Bihar district name checked automatically.
In a weather warning, one wrong figure or place name is a safety problem. IMD Weather Alert MT checks each translation against the English source and tells you when a person needs to review it.
Key features
- Three languages, one model: Hindi, Maithili and Urdu from the same English alert.
- Accurate on weather alerts: made for the fixed wording of IMD bulletins.
- Automatic checks: compares numbers and district names in every translation with the English source.
- Clear review signal: marks each result as verified, or lists exactly what did not match.
- Ready to integrate: a few lines of Python, on GPU or CPU.
Performance
Measured on 1,000 IMD bulletin sentences per language.
| Measure | Hindi | Maithili | Urdu |
|---|---|---|---|
| Numbers exact | 100% | 100% | 100% |
| District names exact | 100% | 100% | 100% |
| BLEU | 75.47 | 55.29 | 78.19 |
| chrF++ | 85.97 | 78.32 | 86.71 |
See it in action
English: Heavy rainfall is most likely to occur at isolated places over Rohtas, Kaimur and Bhojpur districts in next 6 hours.
| Language | Translation |
|---|---|
| Hindi | अगले 6 घंटों के दौरान रोहतास, कैमूर एवं भोजपुर जिलों के अलग-अलग स्थानों पर भारी वर्षा की अत्यधिक संभावना है। |
| Maithili | अगिला 6 घंटाक दौरान रोहतास, कैमूर आ भोजपुर जिलाक एक-दू स्थानपर भारी बरखा होएबाक अत्यधिक संभावना अछि। |
| Urdu | اگلے 6 گھنٹوں کے دوران روہتاس، کیمور اور بھوجپور اضلاع کے الگ الگ مقامات پر بھاری بارش ہونے کا بہت قوی امکان ہے۔ |
English: Frequent cloud-to-ground lightning is very likely to occur at isolated places over Arwal district in next 6 hours.
| Language | Translation |
|---|---|
| Hindi | अगले 6 घंटों के दौरान अरवल जिले के अलग-अलग स्थानों पर बार-बार बादलों से जमीन तक वज्रपात की प्रबल संभावना है। |
| Maithili | अगिला 6 घंटाक दौरान अरवल जिलाक एक-दू स्थानपर बेर-बेर मेघसँ धरती धरि वज्रपात होएबाक प्रबल संभावना अछि। |
| Urdu | اگلے 6 گھنٹوں کے دوران ارول ضلع کے الگ الگ مقامات پر بار بار بادلوں سے زمین تک آسمانی بجلی گرنے کا قوی امکان ہے۔ |
Get started
1. Request access. Click the access button on this page and state your purpose of use. Requests are approved automatically.
2. Install. Log in with a token that has read access, then download the model and its requirements:
hf auth login
hf download helloboyn/imd-tri --local-dir imd-tri
pip install -r imd-tri/requirements.txt
3. Translate. Run this from the folder that contains imd-tri:
import sys
sys.path.insert(0, "imd-tri")
from imd_translate import IMDTranslator
translator = IMDTranslator("imd-tri")
alert = ["Heavy rainfall is very likely at isolated places over Patna district in next 24 hours."]
for lang in ["hin_Deva", "mai_Deva", "urd_Arab"]:
for r in translator.translate(alert, tgt_lang=lang):
print(lang, r["translation"], r["verified"], r["problems"])
tgt_lang |
Language |
|---|---|
hin_Deva |
Hindi |
mai_Deva |
Maithili |
urd_Arab |
Urdu |
Verified results
"verified": Truemeans the numbers and district names match the English source."verified": Falsemeans a mismatch remains, and"problems"lists it. A person must review this translation before it is sent out.
Verification covers numbers and district names only, so human review is still recommended for public warnings.
Scope
- Designed for weather bulletins and alerts, not general-purpose translation.
- Translates English to Hindi, Maithili and Urdu.
- District checks cover the 38 districts of Bihar.
- Downloads last month
- -