resolve copilot comments

Signed-off-by: Will Guo <willg@nvidia.com>

Author: willg-nv

modelopt/onnx/quantization/autotune/common.py

@@ -447,8 +447,11 @@ def from_dict(
     def __str__(self) -> str:
         """String representation for debugging."""
         best_latency = self.best_scheme.latency_ms if self.best_scheme else 0.0
+        pattern_str = self.pattern_signature[:40] + (
+            "..." if len(self.pattern_signature) > 40 else ""
+        )
         return (
-            f"PatternSchemes(pattern='{self.pattern_signature[:40]}...', "
+            f"PatternSchemes(pattern='{pattern_str}', "
             f"schemes={self.num_schemes}, best_latency={best_latency:.3f}ms)"
         )
 
@@ -516,19 +519,22 @@ def add_pattern_schemes(self, pattern_schemes: PatternSchemes) -> None:
             for scheme in sorted_schemes:
                 # Check if this scheme is too similar to any already-filtered scheme
                 too_similar = False
+                existing_to_remove = None  # at most one; remove after inner loop
                 for existing_scheme in filtered_schemes:
                     distance = scheme.distance(existing_scheme)
                     if distance < self.minimum_distance:
                         # Schemes are too similar, keep the better one
                         if scheme.latency_ms < existing_scheme.latency_ms:
-                            # New scheme is better, remove existing and add new
-                            filtered_schemes.remove(existing_scheme)
+                            # New scheme is better; mark existing for removal
+                            existing_to_remove = existing_scheme
                             break
                         else:
                             # Existing scheme is better, skip new one
                             too_similar = True
                             break
 
+                if existing_to_remove is not None:
+                    filtered_schemes.remove(existing_to_remove)
                 if not too_similar:
                     filtered_schemes.append(scheme)
 

modelopt/onnx/quantization/autotune/workflows.py

@@ -84,7 +84,7 @@ def init_benchmark_instance(
     timing_cache_file: str | None = None,
     warmup_runs: int = 5,
     timing_runs: int = 20,
-    trtexec_args: str | None = None,
+    trtexec_args: list[str] | None = None,
 ):
     """Initialize global TensorRT benchmark instance for model performance measurement.
 

tests/gpu/onnx/quantization/autotune/test_workflow.py

@@ -14,6 +14,7 @@
 # limitations under the License.
 
 import os
+import shutil
 import tempfile
 from pathlib import Path
 
@@ -48,7 +49,7 @@ def test_export_quantized_model(use_trtexec, simple_conv_model):
     # Save baseline model
     onnx.save(simple_conv_model, baseline_model_path)
 
-    output_dir = baseline_model_path.strip(".onnx")
+    output_dir = baseline_model_path.replace(".onnx", "")
     output_path = output_dir + ".quant.onnx"
 
     try:
@@ -75,3 +76,7 @@ def test_export_quantized_model(use_trtexec, simple_conv_model):
     finally:
         if os.path.exists(output_path):
             os.unlink(output_path)
+        if os.path.exists(baseline_model_path):
+            os.unlink(baseline_model_path)
+        if os.path.isdir(output_dir):
+            shutil.rmtree(output_dir)