unify arguments format by using whitespace. (#50)

Co-authored-by: Guoshuai Zhao <guzhao@microsoft.com>
Co-authored-by: Yifan Xiong <yifan.xiong@microsoft.com>

examples/benchmarks/matmul.py

@@ -12,7 +12,7 @@
 
 if __name__ == '__main__':
     context = BenchmarkRegistry.create_benchmark_context(
-        'matmul', parameters='--num_steps=20', framework=Framework.PYTORCH
+        'matmul', parameters='--num_steps 20', framework=Framework.PYTORCH
     )
 
     benchmark = BenchmarkRegistry.launch_benchmark(context)

examples/benchmarks/pytorch_bert_large.py

@@ -15,7 +15,7 @@
     # Create context for bert-large benchmark and run it for 120 * 2 seconds.
     context = BenchmarkRegistry.create_benchmark_context(
         'bert-large',
-        parameters='--batch_size=1 --duration=120 --seq_len=512 --precision=float32 --run_count=2',
+        parameters='--batch_size 1 --duration 120 --seq_len 8 --precision float32 --run_count 2',
         framework=Framework.PYTORCH
     )
 

examples/benchmarks/sharding_matmul.py

@@ -12,7 +12,7 @@
 
 if __name__ == '__main__':
     context = BenchmarkRegistry.create_benchmark_context(
-        'sharding-matmul', parameters='--num_steps=20', framework=Framework.PYTORCH
+        'sharding-matmul', parameters='--num_steps 20', framework=Framework.PYTORCH
     )
 
     benchmark = BenchmarkRegistry.launch_benchmark(context)

tests/benchmarks/model_benchmarks/test_model_base.py

@@ -101,13 +101,13 @@ def _cal_params_count(self):
         return 200
 
 
-def create_benchmark(params='--num_steps=8'):
+def create_benchmark(params='--num_steps 8'):
     """Register and create benchmark."""
     # Register the FakeModelBenchmark benchmark.
     BenchmarkRegistry.register_benchmark(
         'pytorch-fake-model',
         FakeModelBenchmark,
-        parameters='--hidden_size=2',
+        parameters='--hidden_size 2',
         platform=Platform.CUDA,
     )
     context = BenchmarkRegistry.create_benchmark_context(
@@ -128,13 +128,13 @@ def test_arguments_related_interfaces():
     Benchmark.get_configurable_settings()
     """
     # Positive case for parse_args().
-    benchmark = create_benchmark('--num_steps=9')
+    benchmark = create_benchmark('--num_steps 9')
     benchmark.add_parser_arguments()
     (ret, args, unknown) = benchmark.parse_args()
     assert (ret and args.num_steps == 9)
 
     # Negative case for parse_args() - invalid precision.
-    benchmark = create_benchmark('--num_steps=8 --precision=fp32')
+    benchmark = create_benchmark('--num_steps 8 --precision fp32')
     benchmark.add_parser_arguments()
     (ret, args, unknown) = benchmark.parse_args()
     assert (ret is False)
@@ -169,7 +169,7 @@ def test_arguments_related_interfaces():
 def test_preprocess():
     """Test interface Benchmark._preprocess()."""
     # Positive case for _preprocess().
-    benchmark = create_benchmark('--num_steps=8')
+    benchmark = create_benchmark('--num_steps 8')
     assert (benchmark._preprocess())
     assert (benchmark.return_code == ReturnCode.SUCCESS)
     settings = benchmark.get_configurable_settings()
@@ -199,12 +199,12 @@ def test_preprocess():
     assert (settings == expected_settings)
 
     # Negative case for _preprocess() - invalid precision.
-    benchmark = create_benchmark('--num_steps=8 --precision=fp32')
+    benchmark = create_benchmark('--num_steps 8 --precision fp32')
     assert (benchmark._preprocess() is False)
     assert (benchmark.return_code == ReturnCode.INVALID_ARGUMENT)
 
     # Negative case for _preprocess() - invalid benchmark type.
-    benchmark = create_benchmark('--num_steps=8 --precision=float32')
+    benchmark = create_benchmark('--num_steps 8 --precision float32')
     benchmark._benchmark_type = Platform.CUDA
     assert (benchmark._preprocess() is False)
     assert (benchmark.return_code == ReturnCode.INVALID_BENCHMARK_TYPE)
@@ -224,7 +224,7 @@ def test_train():
     assert (benchmark.serialized_result == expected_result)
 
     # Step time list is empty (simulate training failure).
-    benchmark = create_benchmark('--num_steps=0')
+    benchmark = create_benchmark('--num_steps 0')
     expected_result = (
         '{"name": "pytorch-fake-model", "type": "model", "run_count": 1, "return_code": 0, '
         '"start_time": null, "end_time": null, "raw_data": {}, "result": {}}'
@@ -248,7 +248,7 @@ def test_inference():
     assert (benchmark.serialized_result == expected_result)
 
     # Step time list is empty (simulate inference failure).
-    benchmark = create_benchmark('--num_steps=0')
+    benchmark = create_benchmark('--num_steps 0')
     expected_result = (
         '{"name": "pytorch-fake-model", "type": "model", "run_count": 1, "return_code": 0, '
         '"start_time": null, "end_time": null, "raw_data": {}, "result": {}}'
@@ -295,19 +295,19 @@ def test_benchmark():
     assert (benchmark.serialized_result == expected_serialized_result)
 
     # Negative case for _benchmark() - no supported precision found.
-    benchmark = create_benchmark('--precision=int16')
+    benchmark = create_benchmark('--precision int16')
     assert (benchmark._preprocess())
     assert (benchmark._benchmark() is False)
     assert (benchmark.return_code == ReturnCode.NO_SUPPORTED_PRECISION)
 
     # Negative case for _benchmark() - model train failure, step time list is empty.
-    benchmark = create_benchmark('--num_steps=0')
+    benchmark = create_benchmark('--num_steps 0')
     assert (benchmark._preprocess())
     assert (benchmark._benchmark() is False)
     assert (benchmark.return_code == ReturnCode.MODEL_TRAIN_FAILURE)
 
     # Negative case for _benchmark() - model inference failure, step time list is empty.
-    benchmark = create_benchmark('--model_action=inference --num_steps=0')
+    benchmark = create_benchmark('--model_action inference --num_steps 0')
     assert (benchmark._preprocess())
     assert (benchmark._benchmark() is False)
     assert (benchmark.return_code == ReturnCode.MODEL_INFERENCE_FAILURE)

tests/benchmarks/model_benchmarks/test_pytorch_base.py

@@ -179,7 +179,7 @@ def test_pytorch_base():
     # Launch benchmark with --no_gpu for testing.
     context = BenchmarkRegistry.create_benchmark_context(
         'pytorch-mnist',
-        parameters='--batch_size=32 --num_warmup=8 --num_steps=64 --model_action train inference --no_gpu'
+        parameters='--batch_size 32 --num_warmup 8 --num_steps 64 --model_action train inference --no_gpu'
     )
 
     benchmark = BenchmarkRegistry.launch_benchmark(context)

tests/benchmarks/model_benchmarks/test_pytorch_bert.py

@@ -15,7 +15,7 @@ def test_pytorch_bert_base():
     context = BenchmarkRegistry.create_benchmark_context(
         'bert-base',
         platform=Platform.CUDA,
-        parameters='--batch_size=32 --num_classes=5 --seq_len=512',
+        parameters='--batch_size 32 --num_classes 5 --seq_len 512',
         framework=Framework.PYTORCH
     )
 
@@ -61,7 +61,7 @@ def test_pytorch_bert_large():
     context = BenchmarkRegistry.create_benchmark_context(
         'bert-large',
         platform=Platform.CUDA,
-        parameters='--batch_size=32 --num_classes=5 --seq_len=512',
+        parameters='--batch_size 32 --num_classes 5 --seq_len 512',
         framework=Framework.PYTORCH
     )
 

tests/benchmarks/test_context.py

@@ -8,8 +8,8 @@
 
 def test_benchmark_context():
     """Test BenchmarkContext class."""
-    context = BenchmarkContext('pytorch-bert-large', Platform.CUDA, 'batch_size=8', framework=Framework.PYTORCH)
+    context = BenchmarkContext('pytorch-bert-large', Platform.CUDA, '--batch_size 8', framework=Framework.PYTORCH)
     assert (context.name == 'pytorch-bert-large')
     assert (context.platform == Platform.CUDA)
-    assert (context.parameters == 'batch_size=8')
+    assert (context.parameters == '--batch_size 8')
     assert (context.framework == Framework.PYTORCH)

tests/benchmarks/test_registry.py

@@ -124,12 +124,12 @@ def test_launch_benchmark():
     """Test interface BenchmarkRegistry.launch_benchmark()."""
     # Register benchmarks for testing.
     BenchmarkRegistry.register_benchmark(
-        'accumulation', AccumulationBenchmark, parameters='--upper_bound=5', platform=Platform.CPU
+        'accumulation', AccumulationBenchmark, parameters='--upper_bound 5', platform=Platform.CPU
     )
 
     # Launch benchmark.
     context = BenchmarkRegistry.create_benchmark_context(
-        'accumulation', platform=Platform.CPU, parameters='--lower_bound=1'
+        'accumulation', platform=Platform.CPU, parameters='--lower_bound 1'
     )
 
     benchmark = BenchmarkRegistry.launch_benchmark(context)
@@ -153,7 +153,7 @@ def test_launch_benchmark():
 
     # Launch benchmark with overridden parameters.
     context = BenchmarkRegistry.create_benchmark_context(
-        'accumulation', platform=Platform.CPU, parameters='--lower_bound=1 --upper_bound=4'
+        'accumulation', platform=Platform.CPU, parameters='--lower_bound 1 --upper_bound 4'
     )
     benchmark = BenchmarkRegistry.launch_benchmark(context)
     assert (benchmark)
@@ -176,14 +176,14 @@ def test_launch_benchmark():
 
     # Failed to launch benchmark due to 'benchmark not found'.
     context = BenchmarkRegistry.create_benchmark_context(
-        'accumulation-fail', Platform.CPU, parameters='--lower_bound=1 --upper_bound=4', framework=Framework.PYTORCH
+        'accumulation-fail', Platform.CPU, parameters='--lower_bound 1 --upper_bound 4', framework=Framework.PYTORCH
     )
     benchmark = BenchmarkRegistry.launch_benchmark(context)
     assert (benchmark is None)
 
     # Failed to launch benchmark due to 'unknown arguments'.
     context = BenchmarkRegistry.create_benchmark_context(
-        'accumulation', platform=Platform.CPU, parameters='--lower_bound=1 --test=4'
+        'accumulation', platform=Platform.CPU, parameters='--lower_bound 1 --test 4'
     )
     benchmark = BenchmarkRegistry.launch_benchmark(context)
     assert (benchmark)
@@ -191,7 +191,7 @@ def test_launch_benchmark():
 
     # Failed to launch benchmark due to 'invalid arguments'.
     context = BenchmarkRegistry.create_benchmark_context(
-        'accumulation', platform=Platform.CPU, parameters='--lower_bound=1 --upper_bound=x'
+        'accumulation', platform=Platform.CPU, parameters='--lower_bound 1 --upper_bound x'
     )
     benchmark = BenchmarkRegistry.launch_benchmark(context)
     assert (benchmark)