akdeniz27's picture
Update README.md
ef20d8a
|
raw
history blame
912 Bytes
metadata
language: tr
license: mit

Turkish Named Entity Recognition (NER) Quantized Model

This model is the dynamically quantized version of the model (https://akdeniz27/convbert-base-turkish-cased-ner)

How to use:

# First install "optimum[onnxruntime]":
!pip install "optimum[onnxruntime]"

# and import "ORTModelForTokenClassification":
from transformers import AutoTokenizer, pipeline
from optimum.onnxruntime import ORTModelForTokenClassification

model_onnx = ORTModelForTokenClassification.from_pretrained("akdeniz27/convbert-base-turkish-cased-ner-quantized", file_name="model_quantized.onnx")
tokenizer = AutoTokenizer.from_pretrained("akdeniz27/convbert-base-turkish-cased-ner-quantized")
ner = pipeline('ner', model=model, tokenizer=tokenizer, aggregation_strategy="first")
ner("your text here")

Pls refer (https://github.com/akdeniz27/dynamic_quantization) for details of quantization.