forked from huggingface/optimum-intel
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtest_pipelines.py
265 lines (249 loc) · 13.3 KB
/
test_pipelines.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
# Copyright 2024 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest
from tempfile import TemporaryDirectory
import numpy as np
import torch
from parameterized import parameterized
from transformers import AutoTokenizer
from transformers.pipelines import pipeline as transformers_pipeline
from optimum.intel.ipex.modeling_base import (
IPEXModelForAudioClassification,
IPEXModelForCausalLM,
IPEXModelForImageClassification,
IPEXModelForMaskedLM,
IPEXModelForQuestionAnswering,
IPEXModelForSequenceClassification,
IPEXModelForTokenClassification,
)
from optimum.intel.pipelines import pipeline as ipex_pipeline
MODEL_NAMES = {
"albert": "hf-internal-testing/tiny-random-albert",
"beit": "hf-internal-testing/tiny-random-BeitForImageClassification",
"bert": "hf-internal-testing/tiny-random-bert",
"bart": "hf-internal-testing/tiny-random-bart",
"blenderbot-small": "hf-internal-testing/tiny-random-BlenderbotModel",
"blenderbot": "hf-internal-testing/tiny-random-BlenderbotModel",
"bloom": "hf-internal-testing/tiny-random-BloomModel",
"convbert": "hf-internal-testing/tiny-random-ConvBertForSequenceClassification",
"codegen": "hf-internal-testing/tiny-random-CodeGenForCausalLM",
"convnext": "hf-internal-testing/tiny-random-convnext",
"distilbert": "hf-internal-testing/tiny-random-distilbert",
"electra": "hf-internal-testing/tiny-random-electra",
"flaubert": "hf-internal-testing/tiny-random-flaubert",
"gpt_bigcode": "hf-internal-testing/tiny-random-GPTBigCodeModel",
"gpt2": "hf-internal-testing/tiny-random-gpt2",
"gpt_neo": "hf-internal-testing/tiny-random-GPTNeoModel",
"gpt_neox": "hf-internal-testing/tiny-random-GPTNeoXForCausalLM",
"gptj": "hf-internal-testing/tiny-random-GPTJModel",
"levit": "hf-internal-testing/tiny-random-LevitModel",
"llama": "fxmarty/tiny-llama-fast-tokenizer",
"llama2": "Jiqing/tiny_random_llama2",
"marian": "sshleifer/tiny-marian-en-de",
"mbart": "hf-internal-testing/tiny-random-mbart",
"mistral": "echarlaix/tiny-random-mistral",
"mobilenet_v1": "google/mobilenet_v1_0.75_192",
"mobilenet_v2": "hf-internal-testing/tiny-random-MobileNetV2Model",
"mobilevit": "hf-internal-testing/tiny-random-mobilevit",
"mpt": "hf-internal-testing/tiny-random-MptForCausalLM",
"mt5": "stas/mt5-tiny-random",
"opt": "hf-internal-testing/tiny-random-OPTModel",
"phi": "echarlaix/tiny-random-PhiForCausalLM",
"resnet": "hf-internal-testing/tiny-random-resnet",
"roberta": "hf-internal-testing/tiny-random-roberta",
"roformer": "hf-internal-testing/tiny-random-roformer",
"squeezebert": "hf-internal-testing/tiny-random-squeezebert",
"t5": "hf-internal-testing/tiny-random-t5",
"unispeech": "hf-internal-testing/tiny-random-unispeech",
"vit": "hf-internal-testing/tiny-random-vit",
"wav2vec2": "anton-l/wav2vec2-random-tiny-classifier",
"xlm": "hf-internal-testing/tiny-random-xlm",
}
class PipelinesIntegrationTest(unittest.TestCase):
COMMON_SUPPORTED_ARCHITECTURES = (
"albert",
"bert",
"distilbert",
"electra",
"flaubert",
"roberta",
"roformer",
"squeezebert",
"xlm",
)
TEXT_GENERATION_SUPPORTED_ARCHITECTURES = (
"bart",
"gpt_bigcode",
"blenderbot",
"blenderbot-small",
"bloom",
"codegen",
"gpt2",
"gpt_neo",
"gpt_neox",
"llama",
"llama2",
"mistral",
"mpt",
"opt",
)
QUESTION_ANSWERING_SUPPORTED_ARCHITECTURES = (
"bert",
"distilbert",
"roberta",
)
AUDIO_CLASSIFICATION_SUPPORTED_ARCHITECTURES = (
"unispeech",
"wav2vec2",
)
IMAGE_CLASSIFICATION_SUPPORTED_ARCHITECTURES = (
"beit",
"mobilenet_v1",
"mobilenet_v2",
"mobilevit",
"resnet",
"vit",
)
@parameterized.expand(COMMON_SUPPORTED_ARCHITECTURES)
def test_token_classification_pipeline_inference(self, model_arch):
model_id = MODEL_NAMES[model_arch]
transformers_generator = transformers_pipeline("token-classification", model_id)
ipex_generator = ipex_pipeline("token-classification", model_id, accelerator="ipex")
inputs = "Hello I'm Omar and I live in Zürich."
with torch.inference_mode():
transformers_output = transformers_generator(inputs)
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertEqual(len(transformers_output), len(ipex_output))
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForTokenClassification))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
for i in range(len(transformers_output)):
self.assertAlmostEqual(transformers_output[i]["score"], ipex_output[i]["score"], delta=1e-4)
@parameterized.expand(COMMON_SUPPORTED_ARCHITECTURES)
def test_sequence_classification_pipeline_inference(self, model_arch):
model_id = MODEL_NAMES[model_arch]
transformers_generator = transformers_pipeline("text-classification", model_id)
ipex_generator = ipex_pipeline("text-classification", model_id, accelerator="ipex")
inputs = "This restaurant is awesome"
with torch.inference_mode():
transformers_output = transformers_generator(inputs)
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForSequenceClassification))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
self.assertEqual(transformers_output[0]["label"], ipex_output[0]["label"])
self.assertAlmostEqual(transformers_output[0]["score"], ipex_output[0]["score"], delta=1e-4)
@parameterized.expand(COMMON_SUPPORTED_ARCHITECTURES)
def test_fill_mask_pipeline_inference(self, model_arch):
model_id = MODEL_NAMES[model_arch]
inputs = "The Milky Way is a <mask> galaxy."
transformers_generator = transformers_pipeline("fill-mask", model_id)
ipex_generator = ipex_pipeline("fill-mask", model_id, accelerator="ipex")
mask_token = transformers_generator.tokenizer.mask_token
inputs = inputs.replace("<mask>", mask_token)
with torch.inference_mode():
transformers_output = transformers_generator(inputs)
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertEqual(len(transformers_output), len(ipex_output))
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForMaskedLM))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
for i in range(len(transformers_output)):
self.assertEqual(transformers_output[i]["token"], ipex_output[i]["token"])
self.assertAlmostEqual(transformers_output[i]["score"], ipex_output[i]["score"], delta=1e-4)
@parameterized.expand(TEXT_GENERATION_SUPPORTED_ARCHITECTURES)
def test_text_generation_pipeline_inference(self, model_arch):
model_id = MODEL_NAMES[model_arch]
transformers_generator = transformers_pipeline("text-generation", model_id)
ipex_generator = ipex_pipeline("text-generation", model_id, accelerator="ipex")
inputs = "Describe a real-world application of AI."
with torch.inference_mode():
transformers_output = transformers_generator(inputs)
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForCausalLM))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
self.assertEqual(transformers_output[0]["generated_text"], ipex_output[0]["generated_text"])
@parameterized.expand(QUESTION_ANSWERING_SUPPORTED_ARCHITECTURES)
def test_question_answering_pipeline_inference(self, model_arch):
model_id = MODEL_NAMES[model_arch]
transformers_generator = transformers_pipeline("question-answering", model_id)
ipex_generator = ipex_pipeline("question-answering", model_id, accelerator="ipex")
question = "How many programming languages does BLOOM support?"
context = "BLOOM has 176 billion parameters and can generate text in 46 languages natural languages and 13 programming languages."
with torch.inference_mode():
transformers_output = transformers_generator(question=question, context=context)
with torch.inference_mode():
ipex_output = ipex_generator(question=question, context=context)
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForQuestionAnswering))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
self.assertAlmostEqual(transformers_output["score"], ipex_output["score"], delta=1e-4)
self.assertEqual(transformers_output["start"], ipex_output["start"])
self.assertEqual(transformers_output["end"], ipex_output["end"])
@parameterized.expand(AUDIO_CLASSIFICATION_SUPPORTED_ARCHITECTURES)
def test_audio_classification_pipeline_inference(self, model_arch):
model_id = MODEL_NAMES[model_arch]
transformers_generator = transformers_pipeline("audio-classification", model_id)
ipex_generator = ipex_pipeline("audio-classification", model_id, accelerator="ipex")
inputs = [np.random.random(16000)]
with torch.inference_mode():
transformers_output = transformers_generator(inputs)
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForAudioClassification))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
self.assertAlmostEqual(transformers_output[0][0]["score"], ipex_output[0][0]["score"], delta=1e-2)
self.assertAlmostEqual(transformers_output[0][1]["score"], ipex_output[0][1]["score"], delta=1e-2)
@parameterized.expand(IMAGE_CLASSIFICATION_SUPPORTED_ARCHITECTURES)
def test_image_classification_pipeline_inference(self, model_arch):
model_id = MODEL_NAMES[model_arch]
transformers_generator = transformers_pipeline("image-classification", model_id)
ipex_generator = ipex_pipeline("image-classification", model_id, accelerator="ipex")
inputs = "http://images.cocodataset.org/val2017/000000039769.jpg"
with torch.inference_mode():
transformers_output = transformers_generator(inputs)
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertEqual(len(transformers_output), len(ipex_output))
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForImageClassification))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
for i in range(len(transformers_output)):
self.assertEqual(transformers_output[i]["label"], ipex_output[i]["label"])
self.assertAlmostEqual(transformers_output[i]["score"], ipex_output[i]["score"], delta=1e-4)
@parameterized.expand(COMMON_SUPPORTED_ARCHITECTURES)
def test_pipeline_load_from_ipex_model(self, model_arch):
model_id = MODEL_NAMES[model_arch]
model = IPEXModelForSequenceClassification.from_pretrained(model_id, export=True)
tokenizer = AutoTokenizer.from_pretrained(model_id)
ipex_generator = ipex_pipeline("text-classification", model, tokenizer=tokenizer, accelerator="ipex")
inputs = "This restaurant is awesome"
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForSequenceClassification))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
self.assertGreaterEqual(ipex_output[0]["score"], 0.0)
@parameterized.expand(COMMON_SUPPORTED_ARCHITECTURES)
def test_pipeline_load_from_jit_model(self, model_arch):
model_id = MODEL_NAMES[model_arch]
model = IPEXModelForSequenceClassification.from_pretrained(model_id, export=True)
save_dir = TemporaryDirectory().name
model.save_pretrained(save_dir)
tokenizer = AutoTokenizer.from_pretrained(model_id)
ipex_generator = ipex_pipeline("text-classification", save_dir, tokenizer=tokenizer, accelerator="ipex")
inputs = "This restaurant is awesome"
with torch.inference_mode():
ipex_output = ipex_generator(inputs)
self.assertTrue(isinstance(ipex_generator.model, IPEXModelForSequenceClassification))
self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
self.assertGreaterEqual(ipex_output[0]["score"], 0.0)