Merge remote-tracking branch 'origin/bench-gpu-profiling' into model_shapes_config

Author: jainapurva

benchmarks/microbenchmarks/benchmark_inference.py

@@ -99,9 +99,10 @@ def run(config: BenchmarkConfig) -> BenchmarkResult:
                 result.profiler_json_path = generate_model_profile(
                     m_copy, input_data, config.profiler_file_name
                 )
-            except Exception:
-                print(f"Error running profiler for {config.name}")
+            except Exception as e:
+                print(f"Error running profiler for {config.name} with error: {e}")
+
         return result
-    except Exception:
-        print(f"Error in benchmark run: {config.name}")
-        return
+    except Exception as e:
+        print(f"Error in benchmark run: {config.name} with error: {e}")
+        return None

benchmarks/microbenchmarks/test/benchmark_config.yml

@@ -5,7 +5,8 @@ quantization_config_recipe_names:
   - "int8wo"
   - "int8dq"
   - "float8dq"
-sparsity_config_recipe_names:
+  - "float8wo"
+# sparsity_config_recipe_names:
   # Will run a baseline inference for model by default, without sparsity for comparison
   - "semi-sparse"
   - "block"

benchmarks/microbenchmarks/utils.py

@@ -77,7 +77,7 @@ def generate_model_profile(model, input_data, profile_file_path):
         record_shapes=True,
         with_stack=True,
         profile_memory=True,
-        with_flops=True,  # Excperiemntal; might be unreliable for some layers
+        with_flops=True,  # Experimental; might be unreliable for some layers
     )
 
     # Warm up