Revert "Speed up GenAi-Perf's help call (#669)" (#670)

Author: the-david-oy

src/c++/perf_analyzer/genai-perf/genai_perf/main.py

@@ -30,33 +30,17 @@
 import traceback
 from argparse import Namespace
 from pathlib import Path
-from typing import TYPE_CHECKING
 
 import genai_perf.logging as logging
+from genai_perf import parser
+from genai_perf.constants import DEFAULT_PARQUET_FILE
 from genai_perf.exceptions import GenAIPerfException
-
-# Import heavy modules to make type checker happy
-if TYPE_CHECKING:
-    from genai_perf import parser
-    from genai_perf.constants import DEFAULT_PARQUET_FILE
-    from genai_perf.export_data.json_exporter import JsonExporter
-    from genai_perf.llm_inputs.llm_inputs import LlmInputs
-    from genai_perf.llm_metrics import LLMProfileDataParser
-    from genai_perf.plots.plot_config_parser import PlotConfigParser
-    from genai_perf.plots.plot_manager import PlotManager
-    from genai_perf.tokenizer import Tokenizer, get_tokenizer
-
-
-def import_heavy_modules():
-    global parser, DEFAULT_PARQUET_FILE, JsonExporter, LlmInputs, LLMProfileDataParser, PlotConfigParser, PlotManager, get_tokenizer
-    from genai_perf import parser
-    from genai_perf.constants import DEFAULT_PARQUET_FILE
-    from genai_perf.export_data.json_exporter import JsonExporter
-    from genai_perf.llm_inputs.llm_inputs import LlmInputs
-    from genai_perf.llm_metrics import LLMProfileDataParser
-    from genai_perf.plots.plot_config_parser import PlotConfigParser
-    from genai_perf.plots.plot_manager import PlotManager
-    from genai_perf.tokenizer import get_tokenizer
+from genai_perf.export_data.json_exporter import JsonExporter
+from genai_perf.llm_inputs.llm_inputs import LlmInputs
+from genai_perf.llm_metrics import LLMProfileDataParser
+from genai_perf.plots.plot_config_parser import PlotConfigParser
+from genai_perf.plots.plot_manager import PlotManager
+from genai_perf.tokenizer import Tokenizer, get_tokenizer
 
 
 def create_artifacts_dirs(args: Namespace) -> None:
@@ -66,9 +50,8 @@ def create_artifacts_dirs(args: Namespace) -> None:
     os.makedirs(plot_dir, exist_ok=True)
 
 
-def generate_inputs(args: Namespace, tokenizer: "Tokenizer") -> None:
+def generate_inputs(args: Namespace, tokenizer: Tokenizer) -> None:
     # TODO (TMA-1759): review if add_model_name is always true
-    import_heavy_modules()
     input_filename = Path(args.input_file.name) if args.input_file else None
     add_model_name = True
     try:
@@ -99,17 +82,14 @@ def generate_inputs(args: Namespace, tokenizer: "Tokenizer") -> None:
     )
 
 
-def calculate_metrics(
-    args: Namespace, tokenizer: "Tokenizer"
-) -> "LLMProfileDataParser":
+def calculate_metrics(args: Namespace, tokenizer: Tokenizer) -> LLMProfileDataParser:
     return LLMProfileDataParser(
         filename=args.profile_export_file,
         tokenizer=tokenizer,
     )
 
 
-def report_output(data_parser: "LLMProfileDataParser", args: Namespace) -> None:
-    import_heavy_modules()
+def report_output(data_parser: LLMProfileDataParser, args: Namespace) -> None:
     if args.concurrency:
         infer_mode = "concurrency"
         load_level = f"{args.concurrency}"
@@ -152,12 +132,10 @@ def run():
     try:
         # TMA-1900: refactor CLI handler
         logging.init_logging()
-        import_heavy_modules()
         args, extra_args = parser.parse_args()
         if args.subcommand == "compare":
             args.func(args)
         else:
-            import_heavy_modules()
             create_artifacts_dirs(args)
             tokenizer = get_tokenizer(args.tokenizer)
             generate_inputs(args, tokenizer)
@@ -169,10 +147,6 @@ def run():
 
 
 def main():
-    # Check if help is requested early
-    if any(arg in sys.argv for arg in ("--help", "-h")):
-        return 0
-
     # Interactive use will catch exceptions and log formatted errors rather than
     # tracebacks.
     try:

src/c++/perf_analyzer/genai-perf/genai_perf/tokenizer.py

@@ -14,16 +14,21 @@
 
 import contextlib
 import io
-from typing import TYPE_CHECKING, Union
+from typing import Union
 
 from genai_perf.exceptions import GenAIPerfException
 
-Tokenizer = Union["PreTrainedTokenizer", "PreTrainedTokenizerFast"]
-DEFAULT_TOKENIZER = "hf-internal-testing/llama-tokenizer"
+# Silence tokenizer warning on import
+with contextlib.redirect_stdout(io.StringIO()) as stdout, contextlib.redirect_stderr(
+    io.StringIO()
+) as stderr:
+    from transformers import AutoTokenizer, PreTrainedTokenizer, PreTrainedTokenizerFast
+    from transformers import logging as token_logger
 
+    token_logger.set_verbosity_error()
 
-if TYPE_CHECKING:
-    from transformers import PreTrainedTokenizer, PreTrainedTokenizerFast
+Tokenizer = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
+DEFAULT_TOKENIZER = "hf-internal-testing/llama-tokenizer"
 
 
 def get_tokenizer(
@@ -37,10 +42,6 @@ def get_tokenizer(
         with contextlib.redirect_stdout(
             io.StringIO()
         ) as stdout, contextlib.redirect_stderr(io.StringIO()) as stderr:
-            from transformers import AutoTokenizer
-            from transformers import logging as token_logger
-
-            token_logger.set_verbosity_error()
             tokenizer = AutoTokenizer.from_pretrained(tokenizer_model)
     except Exception as e:
         raise GenAIPerfException(e)