ModelBenchmarks - Fix early stop logic due to num_steps. (#522)

**Description**
Model benchmarks can stop due to `num_steps` or `duration` config which
will take effect when the value is set greater than 0.
If both are set greater than 0, the earliest condition reached will
work.

docs/superbench-config.mdx

@@ -344,10 +344,10 @@ There have four common parameters for all benchmarks:
 
 For Model-Benchmark, there have some parameters that can control the elapsed time.
 * duration: the elapsed time of benchmark in seconds.
-* num_warmup: the number of warmup step.
+* num_warmup: the number of warmup step, should be positive integer.
 * num_steps: the number of test step.
 
-If `duration > 0` and `num_warmup + num_steps > 0`, then benchmark will take the least as the elapsed time. Otherwise only one of them will take effect.
+If `duration > 0` and `num_steps > 0`, then benchmark will take the least as the elapsed time. Otherwise only one of them will take effect.
 
 ## `Mode` Schema
 

superbench/benchmarks/model_benchmarks/model_base.py

@@ -204,6 +204,11 @@ def _preprocess(self):
             )
         )
 
+        if self._args.num_warmup < 0:
+            logger.error('num_warmup should be positive integer, while {} is set.'.format(self._args.num_warmup))
+            self._result.set_return_code(ReturnCode.INVALID_ARGUMENT)
+            return False
+
         if not self._init_distributed_setting():
             self._result.set_return_code(ReturnCode.DISTRIBUTED_SETTING_INIT_FAILURE)
             return False
@@ -374,7 +379,7 @@ def _is_finished(self, curr_step, curr_time):
 
         if (
             (self._args.duration > 0 and (curr_time - self._sub_benchmark_start_time) >= self._args.duration)
-            or (total_steps > 0 and curr_step >= total_steps)
+            or (self._args.num_steps > 0 and curr_step >= total_steps)
         ):
             return True
 

tests/benchmarks/model_benchmarks/test_model_base.py

@@ -20,6 +20,7 @@ def __init__(self, name, parameters=''):
         """
         super().__init__(name, parameters)
         self._supported_precision = [Precision.FLOAT32, Precision.FLOAT16]
+        self._sub_benchmark_start_time = 0
 
     def add_parser_arguments(self):
         """Add the specified arguments."""
@@ -377,3 +378,37 @@ def test_check_result_format():
     # Negative case for __check_raw_data() - invalid benchmark result.
     assert (benchmark._Benchmark__check_result_format() is False)
     assert (benchmark.return_code == ReturnCode.INVALID_BENCHMARK_RESULT)
+
+
+def test_is_finished():
+    """Test interface Benchmark._is_finished()."""
+    # Only step takes effect, benchmarking finish due to step.
+    benchmark = create_benchmark('--num_warmup 32 --num_steps 128 --duration 0')
+    benchmark._preprocess()
+    end_time = 2
+    curr_step = 50
+    assert (benchmark._is_finished(curr_step, end_time) is False)
+    curr_step = 160
+    assert (benchmark._is_finished(curr_step, end_time))
+
+    # Only duration takes effect, benchmarking finish due to duration.
+    benchmark = create_benchmark('--num_warmup 32 --num_steps 0 --duration 10')
+    benchmark._preprocess()
+    benchmark._sub_benchmark_start_time = 0
+    curr_step = 50
+    end_time = 1
+    assert (benchmark._is_finished(curr_step, end_time) is False)
+    end_time = 10
+    assert (benchmark._is_finished(curr_step, end_time))
+
+    # Both step and duration take effect.
+    benchmark = create_benchmark('--num_warmup 32 --num_steps 128 --duration 10')
+    benchmark._preprocess()
+    # Benchmarking finish due to step.
+    curr_step = 160
+    end_time = 2
+    assert (benchmark._is_finished(curr_step, end_time))
+    # Benchmarking finish due to duration.
+    curr_step = 50
+    end_time = 10
+    assert (benchmark._is_finished(curr_step, end_time))