sravan1817/indicbertv2-ner-te-kn

Multilingual IndicBERTv2 NER Model (Telugu + Kannada )

This model is fine-tuned with limited data(1000 samples from both telugu and kannada combinely) for Named Entity Recognition:

• IndicBERTv2 Backbone
• Languages: Telugu, Kannada
• Tags: PER, LOC, ORG, MISC (BIO format)

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

# Load model
model_name = "sravan1817/indicbertv2-ner-te-kn"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)

# Build NER pipeline
ner = pipeline(
    "ner",
    model=model,
    tokenizer=tokenizer,
    aggregation_strategy="average"   # groups subwords into full words
)

# Example Telugu text
text_te = "విరాట్ కోహ్లీ భారత క్రికెట్ జట్టు మాజీ కెప్టెన్."

# Example Kannada text
text_kn = "ಬೆಂಗಳೂರು ಕರ್ನಾಟಕದ ರಾಜಧಾನಿ ನಗರ."

# Run predictions
print("Telugu Prediction:")
print(ner(text_te))

print("\nKannada Prediction:")
print(ner(text_kn))