update

superbench/benchmarks/model_benchmarks/model_base.py

@@ -443,3 +443,15 @@ class ModelBenchmark(Benchmark):
         """Print environments or dependencies information."""
         # TODO: will implement it when add real benchmarks in the future.
         pass
+
+    def _log_step_time(self, curr_step, precision, start, end):
+        """Log step time into stdout regularly.
+
+        Args:
+            curr_step (int): the index of the current step
+            precision (Precision): precision of model and input data, such as float32, float16.
+            start (float): the start timestamp of the current step
+            end (float): the end timestamp of the current step
+        """
+        if self._args.log_every_steps and curr_step % self._args.log_every_steps == 0:
+            print(f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}')

superbench/benchmarks/model_benchmarks/pytorch_bert.py

@@ -148,9 +148,7 @@ class PytorchBERT(PytorchBase):
                 self._optimizer.step()
                 end = self._timer()
                 curr_step += 1
-                if self._args.log_every_steps:
-                    if curr_step % self._args.log_every_steps == 0:
-                        print(f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}')
+                self._log_step_time(curr_step, precision, start, end)
                 if curr_step > self._args.num_warmup:
                     # Save the step time of every training/inference step, unit is millisecond.
                     duration.append((end - start) * 1000)
@@ -179,11 +177,7 @@ class PytorchBERT(PytorchBase):
                     self._model(sample)
                     end = self._timer()
                     curr_step += 1
-                    if self._args.log_every_steps:
-                        if curr_step % self._args.log_every_steps == 0:
-                            print(
-                                f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}'
-                            )
+                    self._log_step_time(curr_step, precision, start, end)
                     if curr_step > self._args.num_warmup:
                         # Save the step time of every training/inference step, unit is millisecond.
                         duration.append((end - start) * 1000)

superbench/benchmarks/model_benchmarks/pytorch_cnn.py

@@ -111,9 +111,7 @@ class PytorchCNN(PytorchBase):
                 self._optimizer.step()
                 end = self._timer()
                 curr_step += 1
-                if self._args.log_every_steps:
-                    if curr_step % self._args.log_every_steps == 0:
-                        print(f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}')
+                self._log_step_time(curr_step, precision, start, end)
                 if curr_step > self._args.num_warmup:
                     # Save the step time of every training/inference step, unit is millisecond.
                     duration.append((end - start) * 1000)
@@ -143,11 +141,7 @@ class PytorchCNN(PytorchBase):
                     self._model(sample)
                     end = self._timer()
                     curr_step += 1
-                    if self._args.log_every_steps:
-                        if curr_step % self._args.log_every_steps == 0:
-                            print(
-                                f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}'
-                            )
+                    self._log_step_time(curr_step, precision, start, end)
                     if curr_step > self._args.num_warmup:
                         # Save the step time of every training/inference step, unit is millisecond.
                         duration.append((end - start) * 1000)

superbench/benchmarks/model_benchmarks/pytorch_gpt2.py

@@ -142,9 +142,7 @@ class PytorchGPT2(PytorchBase):
                 self._optimizer.step()
                 end = self._timer()
                 curr_step += 1
-                if self._args.log_every_steps:
-                    if curr_step % self._args.log_every_steps == 0:
-                        print(f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}')
+                self._log_step_time(curr_step, precision, start, end)
                 if curr_step > self._args.num_warmup:
                     # Save the step time of every training/inference step, unit is millisecond.
                     duration.append((end - start) * 1000)
@@ -173,11 +171,7 @@ class PytorchGPT2(PytorchBase):
                     self._model(sample)
                     end = self._timer()
                     curr_step += 1
-                    if self._args.log_every_steps:
-                        if curr_step % self._args.log_every_steps == 0:
-                            print(
-                                f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}'
-                            )
+                    self._log_step_time(curr_step, precision, start, end)
                     if curr_step > self._args.num_warmup:
                         # Save the step time of every training/inference step, unit is millisecond.
                         duration.append((end - start) * 1000)

superbench/benchmarks/model_benchmarks/pytorch_lstm.py

@@ -151,9 +151,7 @@ class PytorchLSTM(PytorchBase):
                 self._optimizer.step()
                 end = self._timer()
                 curr_step += 1
-                if self._args.log_every_steps:
-                    if curr_step % self._args.log_every_steps == 0:
-                        print(f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}')
+                self._log_step_time(curr_step, precision, start, end)
                 if curr_step > self._args.num_warmup:
                     # Save the step time of every training/inference step, unit is millisecond.
                     duration.append((end - start) * 1000)
@@ -183,11 +181,7 @@ class PytorchLSTM(PytorchBase):
                     self._model(sample)
                     end = self._timer()
                     curr_step += 1
-                    if self._args.log_every_steps:
-                        if curr_step % self._args.log_every_steps == 0:
-                            print(
-                                f'{self._name} - {precision.value}: step {curr_step}, step time {(end - start) * 1000}'
-                            )
+                    self._log_step_time(curr_step, precision, start, end)
                     if curr_step > self._args.num_warmup:
                         # Save the step time of every training/inference step, unit is millisecond.
                         duration.append((end - start) * 1000)