Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
test_benchmark_tf.py 6.47 KB
Newer Older
Patrick von Platen's avatar
Benchmarks (#4912)  
Patrick von Platen committed
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 
import os
import tempfile
import unittest
from pathlib import Path

from transformers import AutoConfig, is_tf_available

from .utils import require_tf


if is_tf_available():
    import tensorflow as tf
    from transformers import TensorflowBenchmark, TensorflowBenchmarkArguments


@require_tf
class TFBenchmarkTest(unittest.TestCase):
    def check_results_dict_not_empty(self, results):
        for model_result in results.values():
            for batch_size, sequence_length in zip(model_result["bs"], model_result["ss"]):
                result = model_result["result"][batch_size][sequence_length]
                self.assertIsNotNone(result)

    def test_inference_no_configs_eager(self):
        MODEL_ID = "sshleifer/tiny-gpt2"
        benchmark_args = TensorflowBenchmarkArguments(
            models=[MODEL_ID],
            training=False,
            no_inference=False,
            sequence_lengths=[8],
            batch_sizes=[1],
            eager_mode=True,
            no_multi_process=True,
        )
        benchmark = TensorflowBenchmark(benchmark_args)
        results = benchmark.run()
        self.check_results_dict_not_empty(results.time_inference_result)
        self.check_results_dict_not_empty(results.memory_inference_result)

    def test_inference_no_configs_graph(self):
        MODEL_ID = "sshleifer/tiny-gpt2"
        benchmark_args = TensorflowBenchmarkArguments(
            models=[MODEL_ID],
            training=False,
            no_inference=False,
            sequence_lengths=[8],
            batch_sizes=[1],
            no_multi_process=True,
        )
        benchmark = TensorflowBenchmark(benchmark_args)
        results = benchmark.run()
        self.check_results_dict_not_empty(results.time_inference_result)
        self.check_results_dict_not_empty(results.memory_inference_result)

    def test_inference_with_configs_eager(self):
        MODEL_ID = "sshleifer/tiny-gpt2"
        config = AutoConfig.from_pretrained(MODEL_ID)
        benchmark_args = TensorflowBenchmarkArguments(
            models=[MODEL_ID],
            training=False,
            no_inference=False,
            sequence_lengths=[8],
            batch_sizes=[1],
            eager_mode=True,
            no_multi_process=True,
        )
        benchmark = TensorflowBenchmark(benchmark_args, [config])
        results = benchmark.run()
        self.check_results_dict_not_empty(results.time_inference_result)
        self.check_results_dict_not_empty(results.memory_inference_result)

    def test_inference_with_configs_graph(self):
        MODEL_ID = "sshleifer/tiny-gpt2"
        config = AutoConfig.from_pretrained(MODEL_ID)
        benchmark_args = TensorflowBenchmarkArguments(
            models=[MODEL_ID],
            training=False,
            no_inference=False,
            sequence_lengths=[8],
            batch_sizes=[1],
            no_multi_process=True,
        )
        benchmark = TensorflowBenchmark(benchmark_args, [config])
        results = benchmark.run()
        self.check_results_dict_not_empty(results.time_inference_result)
        self.check_results_dict_not_empty(results.memory_inference_result)

    def test_inference_encoder_decoder_with_configs(self):
        MODEL_ID = "patrickvonplaten/t5-tiny-random"
        config = AutoConfig.from_pretrained(MODEL_ID)
        benchmark_args = TensorflowBenchmarkArguments(
            models=[MODEL_ID],
            training=False,
            no_inference=False,
            sequence_lengths=[8],
            batch_sizes=[1],
            no_multi_process=True,
        )
        benchmark = TensorflowBenchmark(benchmark_args, configs=[config])
        results = benchmark.run()
        self.check_results_dict_not_empty(results.time_inference_result)
        self.check_results_dict_not_empty(results.memory_inference_result)

    @unittest.skipIf(is_tf_available() and len(tf.config.list_physical_devices("GPU")) == 0, "Cannot do xla on CPU.")
    def test_inference_no_configs_xla(self):
        MODEL_ID = "sshleifer/tiny-gpt2"
        benchmark_args = TensorflowBenchmarkArguments(
            models=[MODEL_ID],
            training=False,
            no_inference=False,
            sequence_lengths=[8],
            batch_sizes=[1],
            use_xla=True,
            no_multi_process=True,
        )
        benchmark = TensorflowBenchmark(benchmark_args)
        results = benchmark.run()
        self.check_results_dict_not_empty(results.time_inference_result)
        self.check_results_dict_not_empty(results.memory_inference_result)

    def test_save_csv_files(self):
        MODEL_ID = "sshleifer/tiny-gpt2"
        with tempfile.TemporaryDirectory() as tmp_dir:
            benchmark_args = TensorflowBenchmarkArguments(
                models=[MODEL_ID],
                no_inference=False,
                save_to_csv=True,
                sequence_lengths=[8],
                batch_sizes=[1],
                inference_time_csv_file=os.path.join(tmp_dir, "inf_time.csv"),
                inference_memory_csv_file=os.path.join(tmp_dir, "inf_mem.csv"),
                env_info_csv_file=os.path.join(tmp_dir, "env.csv"),
                no_multi_process=True,
            )
            benchmark = TensorflowBenchmark(benchmark_args)
            benchmark.run()
            self.assertTrue(Path(os.path.join(tmp_dir, "inf_time.csv")).exists())
            self.assertTrue(Path(os.path.join(tmp_dir, "inf_mem.csv")).exists())
            self.assertTrue(Path(os.path.join(tmp_dir, "env.csv")).exists())

    def test_trace_memory(self):
        MODEL_ID = "sshleifer/tiny-gpt2"

        def _check_summary_is_not_empty(summary):
            self.assertTrue(hasattr(summary, "sequential"))
            self.assertTrue(hasattr(summary, "cumulative"))
            self.assertTrue(hasattr(summary, "current"))
            self.assertTrue(hasattr(summary, "total"))

        with tempfile.TemporaryDirectory() as tmp_dir:
            benchmark_args = TensorflowBenchmarkArguments(
                models=[MODEL_ID],
                no_inference=False,
                sequence_lengths=[8],
                batch_sizes=[1],
                log_filename=os.path.join(tmp_dir, "log.txt"),
                log_print=True,
                trace_memory_line_by_line=True,
                eager_mode=True,
                no_multi_process=True,
            )
            benchmark = TensorflowBenchmark(benchmark_args)
            result = benchmark.run()
            _check_summary_is_not_empty(result.inference_summary)
            self.assertTrue(Path(os.path.join(tmp_dir, "log.txt")).exists())