def test_inference_large_model(self):
model = LukeModel.from_pretrained("studio-ousia/luke-large").eval()
model.to(torch_device)
tokenizer = LukeTokenizer.from_pretrained("studio-ousia/luke-large", task="entity_classification")
text = "Top seed Ana Ivanovic said on Thursday she could hardly believe her luck as a fortuitous netcord helped the new world number one avoid a humiliating second- round exit at Wimbledon ."
span = (39, 42)
encoding = tokenizer(text, entity_spans=[span], add_prefix_space=True, return_tensors="pt")
# move all values to device
for key, value in encoding.items():
encoding[key] = encoding[key].to(torch_device)
outputs = model(**encoding)
# Verify word hidden states
expected_shape = torch.Size((1, 42, 1024))
self.assertEqual(outputs.last_hidden_state.shape, expected_shape)
expected_slice = torch.tensor(
[[0.0133, 0.0865, 0.0095], [0.3093, -0.2576, -0.7418], [-0.1720, -0.2117, -0.2869]]
).to(torch_device)
self.assertTrue(torch.allclose(outputs.last_hidden_state[0, :3, :3], expected_slice, atol=1e-4))
# Verify entity hidden states
expected_shape = torch.Size((1, 1, 1024))
self.assertEqual(outputs.entity_last_hidden_state.shape, expected_shape)
expected_slice = torch.tensor([[0.0466, -0.0106, -0.0179]]).to(torch_device)
self.assertTrue(torch.allclose(outputs.entity_last_hidden_state[0, :3, :3], expected_slice, atol=1e-4))
def test_model_from_pretrained(self):
for model_name in LUKE_PRETRAINED_MODEL_ARCHIVE_LIST:
model = LukeModel.from_pretrained(model_name)
self.assertIsNotNone(model)
# @author Loreto Parisi (loretoparisi at gmail dot com)
# Copyright (c) 2021 Loreto Parisi (loretoparisi at gmail dot com)
import os
from transformers import LukeTokenizer, LukeModel, LukeForEntityPairClassification, LukeForEntitySpanClassification
###
# LUKE models in the following examples will be saved to cache_dir=../../models
#
# studio-ousia/luke-base
# studio-ousia/luke-large-finetuned-tacred
# studio-ousia/luke-large-finetuned-conll-2003
###
model = LukeModel.from_pretrained("studio-ousia/luke-base",
cache_dir=os.getenv("cache_dir",
"../../models"))
tokenizer = LukeTokenizer.from_pretrained("studio-ousia/luke-base",
cache_dir=os.getenv(
"cache_dir", "../../models"))
# Example 1: Computing the contextualized entity representation corresponding to the entity mention "Beyoncé"
text = "Beyoncé lives in Los Angeles."
entity_spans = [(0, 7)
] # character-based entity span corresponding to "Beyoncé"
inputs = tokenizer(text,
entity_spans=entity_spans,
add_prefix_space=True,
return_tensors="pt")
outputs = model(**inputs)
word_last_hidden_state = outputs.last_hidden_state