build: clean up build

Dockerfile

@@ -71,17 +71,10 @@ RUN chmod +x /opt/triton-model-analyzer/nvidia_entrypoint.sh
 RUN chmod +x build_wheel.sh && \
     ./build_wheel.sh perf_analyzer true && \
     rm -f perf_analyzer
-RUN python3 -m pip install nvidia-pyindex && \
-    python3 -m pip install wheels/triton_model_analyzer-*-manylinux*.whl
-# Install other pip packages
-RUN python3 -m pip install coverage
-RUN python3 -m pip install mypy
-RUN python3 -m pip install types-PyYAML
-RUN python3 -m pip install types-requests
-RUN python3 -m pip install types-protobuf
-RUN python3 -m pip install mkdocs
-RUN python3 -m pip install mkdocs-htmlproofer-plugin==0.10.3
-RUN python3 -m pip install yapf==0.32.0
+
+# Install model analyzer and development dependencies
+RUN python3 -m pip install --no-cache-dir wheels/triton_model_analyzer-*-manylinux*.whl && \
+    python3 -m pip install --no-cache-dir -r requirements-dev.txt
 
 RUN apt-get install -y wkhtmltopdf
 

experiments/experiment_config_command_creator.py

@@ -37,23 +37,26 @@ def make_config(data_path, model_name, other_args):
         else:
             checkpoint_dir = f"{data_path}/{model_name}"
 
-        # yapf: disable
         args = [
-            'model-analyzer', 'profile',
-            '--profile-models', model_name,
-            '--model-repository', data_path,
-            '--checkpoint-directory', checkpoint_dir
+            "model-analyzer",
+            "profile",
+            "--profile-models",
+            model_name,
+            "--model-repository",
+            data_path,
+            "--checkpoint-directory",
+            checkpoint_dir,
         ]
         args += other_args
 
-        if '-f' not in args and '--config-file' not in args:
-            args += ['-f', 'path-to-config-file']
+        if "-f" not in args and "--config-file" not in args:
+            args += ["-f", "path-to-config-file"]
             yaml_content = convert_to_bytes("")
         else:
-            index = args.index('-f') if '-f' in args else args.index('--config-file')
+            index = args.index("-f") if "-f" in args else args.index("--config-file")
             yaml_file = args[index + 1]
 
-            with open(yaml_file, 'r') as f:
+            with open(yaml_file, "r") as f:
                 yaml_content = f.read()
                 yaml_content = convert_to_bytes(yaml_content)
 
@@ -65,10 +68,11 @@ def make_config(data_path, model_name, other_args):
         config = ConfigCommandExperiment()
         cli = CLI()
         cli.add_subcommand(
-            cmd='profile',
-            help='Run model inference profiling based on specified CLI or '
-            'config options.',
-            config=config)
+            cmd="profile",
+            help="Run model inference profiling based on specified CLI or "
+            "config options.",
+            config=config,
+        )
         cli.parse()
         mock_config.stop()
 

experiments/experiment_file_writer.py

@@ -56,21 +56,20 @@ def write(self, checkpoint_data, profile_data, radius, min_initialized):
                 )
                 quick_best_measurement = profile_data.get_best_run_config_measurement()
 
-                # yapf: disable
-                writer.writerow({
-                    "overall_num_measurements":
-                        checkpoint_data.get_run_config_measurement_count(),
-                    "overall_best_throughput":
-                        overall_best_measurement.get_non_gpu_metric_value("perf_throughput"),
-                    "quick_num_measurements":
-                        profile_data.get_run_config_measurement_count(),
-                    "missing_num_measurements":
-                        checkpoint_data.get_missing_measurement_count(),
-                    "quick_throughput":
-                        quick_best_measurement.get_non_gpu_metric_value("perf_throughput"),
-                    "radius": radius,
-                    "min_initialized": min_initialized
-                })
-                # yapf: enable
+                writer.writerow(
+                    {
+                        "overall_num_measurements": checkpoint_data.get_run_config_measurement_count(),
+                        "overall_best_throughput": overall_best_measurement.get_non_gpu_metric_value(
+                            "perf_throughput"
+                        ),
+                        "quick_num_measurements": profile_data.get_run_config_measurement_count(),
+                        "missing_num_measurements": checkpoint_data.get_missing_measurement_count(),
+                        "quick_throughput": quick_best_measurement.get_non_gpu_metric_value(
+                            "perf_throughput"
+                        ),
+                        "radius": radius,
+                        "min_initialized": min_initialized,
+                    }
+                )
         except OSError as e:
             raise TritonModelAnalyzerException(e)

experiments/scripts/test_pa.py

@@ -343,15 +343,21 @@ def _add_results(self, config, results):
         self._results.append((config, results))
 
     def _get_cmd(self, config: RunConfigData):
-        # yapf: disable
         cmd = [
-            "/usr/local/bin/perf_analyzer", "-v",
-            "-i", config.protocol,
-            "--measurement-mode", config.measurement_mode,
-            "-m", config.model,
-            "-b", str(config.batch_size),
-            "--concurrency-range", str(config.concurrency),
-            "--max-threads", str(config.max_threads)
+            "/usr/local/bin/perf_analyzer",
+            "-v",
+            "-i",
+            config.protocol,
+            "--measurement-mode",
+            config.measurement_mode,
+            "-m",
+            config.model,
+            "-b",
+            str(config.batch_size),
+            "--concurrency-range",
+            str(config.concurrency),
+            "--max-threads",
+            str(config.max_threads),
         ]
         if config.protocol == "http":
             cmd += ["-u", "localhost:8000"]
@@ -362,7 +368,6 @@ def _get_cmd(self, config: RunConfigData):
             cmd += ["--async"]
         else:
             cmd += ["--sync"]
-        # yapf: enable
         return cmd
 
     def _get_dict_combos(self, config: dict):