Clean up code around load args

Author: tgerdesnv

model_analyzer/config/generate/model_profile_spec.py

@@ -22,6 +22,7 @@
     ConfigModelProfileSpec,
 )
 from model_analyzer.device.gpu_device import GPUDevice
+from model_analyzer.perf_analyzer.perf_config import PerfAnalyzerConfig
 from model_analyzer.triton.client.client import TritonClient
 from model_analyzer.triton.model.model_config import ModelConfig
 
@@ -72,3 +73,12 @@ def supports_dynamic_batching(self) -> bool:
     def is_ensemble(self) -> bool:
         """Returns true if the model is an ensemble"""
         return "ensemble_scheduling" in self._default_model_config
+
+    def is_load_specified(self) -> bool:
+        """
+        Returns true if the model's PA config has specified any of the
+        inference load args (such as concurrency). Else returns false
+        """
+        load_args = PerfAnalyzerConfig.get_inference_load_args()
+        pa_flags = self.perf_analyzer_flags()
+        return any(e in pa_flags for e in load_args)

model_analyzer/config/generate/quick_run_config_generator.py

@@ -512,8 +512,7 @@ def _get_next_perf_analyzer_config(
 
         perf_analyzer_config.update_config_from_profile_config(model_name, self._config)
 
-        # FIXME 1772 -- use new method in perf_config
-        if not "request-intervals" in model.perf_analyzer_flags():
+        if not model.is_load_specified():
             concurrency = self._calculate_concurrency(dimension_values)
             perf_config_params = {"concurrency-range": concurrency}
             perf_analyzer_config.update_config(perf_config_params)

model_analyzer/perf_analyzer/perf_config.py

@@ -96,6 +96,13 @@ class PerfAnalyzerConfig:
         "collect-metrics",
     ]
 
+    # Only one of these args can be sent to PA, as each one controls the inference load in a different way
+    inference_load_args = [
+        "concurrency-range",
+        "request-rate-range",
+        "request-intervals",
+    ]
+
     def __init__(self):
         """
         Construct a PerfAnalyzerConfig
@@ -160,6 +167,16 @@ def additive_keys(cls):
 
         return cls.additive_args[:]
 
+    @classmethod
+    def get_inference_load_args(cls):
+        """
+        Returns
+        -------
+        list of str
+            The Perf Analyzer args that control the inference load
+        """
+        return cls.inference_load_args
+
     def update_config(self, params=None):
         """
         Allows setting values from a params dict

model_analyzer/plots/detailed_plot.py

@@ -22,6 +22,7 @@
 from matplotlib import patches as mpatches
 
 from model_analyzer.constants import LOGGER_NAME
+from model_analyzer.perf_analyzer.perf_config import PerfAnalyzerConfig
 from model_analyzer.record.metrics_manager import MetricsManager
 
 logging.getLogger("matplotlib").setLevel(logging.ERROR)
@@ -119,42 +120,15 @@ def add_run_config_measurement(self, run_config_measurement):
         """
 
         # TODO-TMA-568: This needs to be updated because there will be multiple model configs
-        if (
-            "concurrency-range" in run_config_measurement.model_specific_pa_params()[0]
-            and run_config_measurement.model_specific_pa_params()[0][
-                "concurrency-range"
-            ]
-        ):
-            self._data["concurrency"].append(
-                run_config_measurement.model_specific_pa_params()[0][
-                    "concurrency-range"
-                ]
-            )
-
-        if (
-            "request-rate-range" in run_config_measurement.model_specific_pa_params()[0]
-            and run_config_measurement.model_specific_pa_params()[0][
-                "request-rate-range"
-            ]
-        ):
-            self._data["request_rate"].append(
-                run_config_measurement.model_specific_pa_params()[0][
-                    "request-rate-range"
-                ]
-            )
-
-        # FIXME 1772 -- clean this up??
-        if (
-            "request-intervals" in run_config_measurement.model_specific_pa_params()[0]
-            and run_config_measurement.model_specific_pa_params()[0][
-                "request-intervals"
-            ]
-        ):
-            self._data["request-intervals"].append(
-                run_config_measurement.model_specific_pa_params()[0][
-                    "request-intervals"
-                ]
-            )
+        for load_arg in PerfAnalyzerConfig.get_inference_load_args():
+            if (
+                load_arg in run_config_measurement.model_specific_pa_params()[0]
+                and run_config_measurement.model_specific_pa_params()[0][load_arg]
+            ):
+                data_key = self._get_data_key_from_load_arg(load_arg)
+                self._data[data_key].append(
+                    run_config_measurement.model_specific_pa_params()[0][load_arg]
+                )
 
         self._data["perf_throughput"].append(
             run_config_measurement.get_non_gpu_metric_value(tag="perf_throughput")
@@ -177,9 +151,9 @@ def plot_data(self):
         """
 
         # Update the x-axis plot title
-        if "request-intervals" in self._data and self._data["request-intervals"][0]:
+        if "request_intervals" in self._data and self._data["request_intervals"][0]:
             self._ax_latency.set_xlabel("Request Intervals File")
-            sort_indices_key = "request-intervals"
+            sort_indices_key = "request_intervals"
         elif "request_rate" in self._data and self._data["request_rate"][0]:
             self._ax_latency.set_xlabel("Client Request Rate")
             sort_indices_key = "request_rate"
@@ -274,3 +248,18 @@ def save(self, filepath):
         """
 
         self._fig.savefig(os.path.join(filepath, self._name))
+
+    def _get_data_key_from_load_arg(self, load_arg):
+        """
+        Gets the key into _data corresponding with the input load arg
+
+        For example, the load arg "request-rate-range" has the key "request_rate"
+        """
+        # Check if '-range' exists at the end of the input string and remove it
+        if load_arg.endswith("-range"):
+            load_arg = load_arg[:-6]
+
+        # Replace any '-' with '_' in the remaining string
+        data_key = load_arg.replace("-", "_")
+
+        return data_key