Add comment for silencing and handle extra args

src/c++/perf_analyzer/genai-pa/genai_pa/llm_metrics.py

@@ -34,6 +34,7 @@
 import numpy as np
 from genai_pa.utils import load_json
 
+# Silence tokenizer warning on import
 with contextlib.redirect_stdout(io.StringIO()) as stdout, contextlib.redirect_stderr(
     io.StringIO()
 ) as stderr:

src/c++/perf_analyzer/genai-pa/genai_pa/main.py

@@ -50,6 +50,21 @@
 logger = logging.getLogger(LOGGER_NAME)
 
 
+def handle_pass_through_args(args):
+    if args == None:
+        args = sys.argv
+
+    extra_args = []
+    if "--" in args:
+        index = args.index("--")
+        if index != (len(args) - 1):
+            extra_args = args[index + 1 : -1]
+            args = args[1:index]
+    else:
+        args = args[1:]
+    return args, extra_args
+
+
 def generate_inputs(args):
     LlmInputs.create_openai_llm_inputs(
         args.dataset,
@@ -74,9 +89,10 @@ def report_output(metrics: LLMProfileData):
 # Optional argv used for testing - will default to sys.argv if None.
 def run(argv=None):
     try:
+        argv, extra_args = handle_pass_through_args(argv)
         args = parser.parse_args(argv)
         generate_inputs(args)
-        args.func(args)
+        args.func(args, extra_args)
         metrics = calculate_metrics(args.profile_export_file)
         report_output(metrics)
     except Exception as e:

src/c++/perf_analyzer/genai-pa/genai_pa/parser.py

@@ -73,10 +73,10 @@ def _verify_valid_arg_combination(
 ### Handlers ###
 
 
-def handler(args):
+def handler(args, extra_args):
     from genai_pa.wrapper import Profiler
 
-    Profiler.run(model=args.model, args=args)
+    Profiler.run(model=args.model, args=args, extra_args=extra_args)
 
 
 ### Parsers ###

src/c++/perf_analyzer/genai-pa/genai_pa/wrapper.py

@@ -35,7 +35,7 @@
 
 class Profiler:
     @staticmethod
-    def build_cmd(model, args):
+    def build_cmd(model, args, extra_args):
         skip_args = ["model", "func", "dataset"]
         if hasattr(args, "version"):
             cmd = f"perf_analyzer --version"
@@ -56,14 +56,17 @@ def build_cmd(model, args):
                     else:
                         arg = utils.convert_option_name(arg)
                         cmd += f"--{arg} {value} "
+
+            for arg in extra_args:
+                cmd += f"{arg} "
         # TODO: Once the OpenAI endpoint support is in place in PA core,
         # update the input-data option arg
         # cmd += f"--input-data {DEFAULT_INPUT_DATA_JSON}  -p 10000 -s 99"
         cmd += f"--input-data ./input_data.json -p 10000 -s 99"
         return cmd
 
     @staticmethod
-    def run(model, args=None):
-        cmd = Profiler.build_cmd(model, args)
+    def run(model, args=None, extra_args=None):
+        cmd = Profiler.build_cmd(model, args, extra_args)
         logger.info(f"Running Perf Analyzer : '{cmd}'")
         subprocess.run(cmd, shell=True, check=True)