Use Allennlp for NER programmatically, and test the runtime.

ner_allennlp_test.py

from allennlp.predictors.predictor import Predictor                                                                                                                             
import time                                                                                                                                                                     
                                                                                                                                                                                
model = "https://s3-us-west-2.amazonaws.com/allennlp/models/ner-model-2018.04.26.tar.gz"                                                                                        
                                                                                                                                                                                
print("Loading model...")                                                                                                                                                       
predictor = Predictor.from_path(model)                                                                                                                                          
print("Done loading model.")                                                                                                                                                    

# this is so we can use pre-tokenized text. Just remove the next 3 lines
# if you want the text tokenized first.
from allennlp.data.tokenizers.word_splitter import JustSpacesWordSplitter                                                                                                       
spacestok = JustSpacesWordSplitter()                                                                                                                                            
predictor._tokenizer = spacestok                                                                                                                                                

# pre-tokenized text.
sent = "Stephen Mayhew is a person who lives in the North Pole ."                                                                                                                

# tag 100 times individually.
start = time.time()                                                                                                                                                             
for i in range(100):                                                                                                                                                            
    res = predictor.predict_json({"sentence" : sent})                                                                                                                           
end = time.time()                                                                                                                                                               
print(end - start)                                                                                                                                                              

# create a batch of 100 elements and tag.
start = time.time()                                                                                                                                                             
sents = []                                                                                                                                                                      
for i in range(100):                                                                                                                                                            
    sents.append({"sentence": sent})                                                                                                                                            
res = predictor.predict_batch_json(sents)                                                                                                                                       
end = time.time()                                                                                                                                                               
print(end - start)

# Output (with GPU):
# Loading model...
# Done loading model.
# 19.09427809715271
# 2.3905272483825684