Benchmarks: Add Feature - Set reduce type for current benchmarks' metrics. (#149)

**Description**
Set reduce type for current benchmarks' metrics, including model benchmarks and ShardingMatmul.

superbench/benchmarks/micro_benchmarks/sharding_matmul.py

@@ -22,6 +22,7 @@ from superbench.common.utils import logger
 from superbench.benchmarks import DistributedImpl, DistributedBackend, BenchmarkRegistry, ReturnCode
 from superbench.benchmarks.micro_benchmarks import MicroBenchmark
 from superbench.benchmarks.context import Enum
+from superbench.benchmarks.reducer import ReduceType
 
 
 class ShardingMode(Enum):
@@ -256,7 +257,7 @@ class ShardingMatmul(MicroBenchmark):
                 return False
 
             metric = '{}'.format(mode)
-            if not self._process_numeric_result(metric, elapse_times):
+            if not self._process_numeric_result(metric, elapse_times, reduce_type=ReduceType.MAX):
                 return False
 
             logger.info(

superbench/benchmarks/model_benchmarks/model_base.py

@@ -12,6 +12,7 @@ from superbench.common.utils import logger
 from superbench.benchmarks import Precision, ModelAction, DistributedImpl, DistributedBackend, BenchmarkType, ReturnCode
 from superbench.benchmarks.base import Benchmark
 from superbench.benchmarks.context import Enum
+from superbench.benchmarks.reducer import ReduceType
 
 
 class Optimizer(Enum):
@@ -358,7 +359,7 @@ class ModelBenchmark(Benchmark):
         metric = 'steptime_{}_{}'.format(model_action, precision)
         self._result.add_raw_data(metric, step_times)
         avg = statistics.mean(step_times)
-        self._result.add_result(metric, avg)
+        self._result.add_result(metric, avg, reduce_type=ReduceType.MAX if model_action is ModelAction.TRAIN else None)
 
         # The unit of step time is millisecond, use it to calculate the throughput with the unit samples/sec.
         millisecond_per_second = 1000
@@ -366,7 +367,7 @@ class ModelBenchmark(Benchmark):
         metric = 'throughput_{}_{}'.format(model_action, precision)
         self._result.add_raw_data(metric, throughput)
         avg = statistics.mean(throughput)
-        self._result.add_result(metric, avg)
+        self._result.add_result(metric, avg, reduce_type=ReduceType.MIN if model_action is ModelAction.TRAIN else None)
 
         return True
 

tests/benchmarks/model_benchmarks/test_model_base.py

@@ -220,7 +220,7 @@ def test_train():
         '"steptime_train_float32": [[2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0]], '
         '"throughput_train_float32": [[16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0]]}, '
         '"result": {"steptime_train_float32": [2.0], "throughput_train_float32": [16000.0]}, '
-        '"reduce": {"steptime_train_float32": null, "throughput_train_float32": null}}'
+        '"reduce": {"steptime_train_float32": "max", "throughput_train_float32": "min"}}'
     )
     assert (benchmark._preprocess())
     assert (benchmark._ModelBenchmark__train(Precision.FLOAT32))
@@ -296,8 +296,8 @@ def test_benchmark():
         '"throughput_train_float16": [[16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0, 16000.0]]}, '
         '"result": {"steptime_train_float32": [2.0], "throughput_train_float32": [16000.0], '
         '"steptime_train_float16": [2.0], "throughput_train_float16": [16000.0]}, '
-        '"reduce": {"steptime_train_float32": null, "throughput_train_float32": null, '
-        '"steptime_train_float16": null, "throughput_train_float16": null}}'
+        '"reduce": {"steptime_train_float32": "max", "throughput_train_float32": "min", '
+        '"steptime_train_float16": "max", "throughput_train_float16": "min"}}'
     )
     assert (benchmark.serialized_result == expected_serialized_result)