README.md

metadata

language:
  - en
tags:
  - pytorch
  - ner
  - text generation
  - seq2seq
inference: false
license: mit
datasets:
  - conll2003
metrics:
  - f1

t5-base-qa-ner-conll

Unofficial implementation of InstructionNER. t5-base model tuned on conll2003 dataset.

https://github.com/ovbystrova/InstructionNER

Inference

git clone https://github.com/ovbystrova/InstructionNER 
cd InstructionNER

from instruction_ner.model import Model

model = Model(
    model_path_or_name="olgaduchovny/t5-base-ner-mit-restaurant",
    tokenizer_path_or_name="olgaduchovny/t5-base-mit-restaurant"
)

options = ["LOC", "PER", "ORG", "MISC"]

instruction = "please extract entities and their types from the input sentence, " \
              "all entity types are in options"

text = "Once I visited Sovok in Nizny Novgorod. I had asian wok there. It was the best WOK i ever had"\
        "It was cheap but lemonades cost 5 dollars."

generation_kwargs = {
    "num_beams": 2,
    "max_length": 128
}

pred_spans = model.predict(
    text=text,
    generation_kwargs=generation_kwargs,
    instruction=instruction,
    options=options
)

>>> ('sovok is a Restaurant_Name, Nizny Novgorod is a Location, asian wok is a Dish, cheap is a Price, lemonades is a Dish, 5 dollars is a Price.',
 [(24, 38, 'Location'),
  (46, 55, 'Dish'),
  (100, 105, 'Price'),
  (110, 119, 'Dish'),
  (125, 134, 'Price')])