Metric Export to CSV added

Author: DEKHTIARJonathan

tftrt/examples/benchmark_args.py

@@ -256,6 +256,14 @@ def __init__(self):
             "to the set location in JSON format for further processing."
         )
 
+        self._parser.add_argument(
+            "--export_metrics_csv_path",
+            type=str,
+            default=None,
+            help="If set, the script will export runtime metrics and arguments "
+            "to the set location in CSV format for further processing."
+        )
+
         self._parser.add_argument(
             "--tf_profile_export_path",
             type=str,

tftrt/examples/benchmark_runner.py

@@ -7,6 +7,7 @@
 import abc
 import contextlib
 import copy
+import csv
 import json
 import logging
 import sys
@@ -125,28 +126,70 @@ def _debug_print(self, msg):
 
     def _export_runtime_metrics_to_json(self, metric_dict):
 
-        metric_dict = {
-          # Creating a copy to avoid modifying the original
-          "results": copy.deepcopy(metric_dict),
-          "runtime_arguments": vars(self._args)
-        }
+        try:
 
-        json_path = self._args.export_metrics_json_path
-        if json_path is not None:
-            try:
-                with open(json_path, 'w') as json_f:
-                    json_string = json.dumps(
-                        metric_dict,
-                        default=lambda o: o.__dict__,
-                        sort_keys=True,
-                        indent=4
-                    )
-                    print(json_string, file=json_f)
-            except Exception as e:
-                print(
-                    "[ERROR] Impossible to save JSON File at path: "
-                    f"{json_path}.\nError: {str(e)}"
+            file_path = self._args.export_metrics_json_path
+            if file_path is None:
+                return
+
+            metric_dict = {
+              # Creating a copy to avoid modifying the original
+              "results": copy.deepcopy(metric_dict),
+              "runtime_arguments": vars(self._args)
+            }
+
+            with open(file_path, 'w') as json_f:
+                json_string = json.dumps(
+                    metric_dict,
+                    default=lambda o: o.__dict__,
+                    sort_keys=True,
+                    indent=4
                 )
+                print(json_string, file=json_f)
+
+        except Exception as e:
+            print(f"An exception occured during export to JSON: {e}")
+
+    def _export_runtime_metrics_to_csv(self, metric_dict):
+
+        try:
+
+            file_path = self._args.export_metrics_csv_path
+            if file_path is None:
+                return
+
+            data = {f"metric_{k}": v for k, v in metric_dict.items()}
+
+            args_to_save = [
+                "batch_size",
+                "input_saved_model_dir",
+                "minimum_segment_size",
+                "no_tf32",
+                "precision",
+                "use_dynamic_shape",
+                "use_synthetic_data",
+                "use_tftrt",
+                "use_xla",
+                "use_xla_auto_jit"
+            ]
+-
+            runtime_arguments = vars(self._args)
+            for key in args_to_save:
+                data[f"arg_{key}"] = str(runtime_arguments[key]).split("/")[-1]
+
+            fieldnames = sorted(data.keys())
+
+            if not os.path.isfile(file_path):
+                with open(file_path, 'w') as outcsv:
+                    writer = csv.DictWriter(outcsv, fieldnames=fieldnames, delimiter=',')
+                    writer.writeheader()
+
+            with open(file_path, 'a') as outcsv:
+                writer = csv.DictWriter(outcsv, fieldnames=fieldnames, delimiter=',')
+                writer.writerow(data)
+
+        except Exception as e:
+            print(f"An exception occured during export to CSV: {e}")
 
     def _get_graph_func(self):
         """Retreives a frozen SavedModel and applies TF-TRT
@@ -524,6 +567,7 @@ def timing_metrics(time_arr, log_prefix):
             metrics.update(timing_metrics(memcopy_times, "Data MemCopyHtoD Time"))
 
             self._export_runtime_metrics_to_json(metrics)
+            self._export_runtime_metrics_to_csv(metrics)
 
             def log_value(key, val):
                 if isinstance(val, int):

tftrt/examples/benchmark_utils.py

@@ -49,7 +49,7 @@ def timed_section(msg, activate=True, start_end_mode=True):
         total_time = time.time() - start_time
 
         if start_end_mode:
-            print(f"[END] `{msg}` - Duration: {total_time:.1f}s")
+            print(f"[END] {msg} - Duration: {total_time:.1f}s")
             print("=" * 80, "\n")
         else:
             print(f"{msg:18s}: {total_time:.4f}s")