YOLO v8: check_export_not_strict

Author: daniil-lyakhov

aa_torch_fx.py

@@ -403,12 +403,12 @@ def process_model(model_name: str):
     ##############################################################
     # Process PT Quantize
     ##############################################################
-    # fx_2_ov_quantization(pt_model, example_input, output_dir, result, val_loader, shape_input)
+    fx_2_ov_quantization(pt_model, example_input, output_dir, result, val_loader, shape_input)
 
     ##############################################################
     # Process NNCF FX Quantize
     ##############################################################
-    nncf_fx_2_ov_quantization(pt_model, example_input, output_dir, result, val_loader, shape_input)
+    # nncf_fx_2_ov_quantization(pt_model, example_input, output_dir, result, val_loader, shape_input)
 
     ##############################################################
     # Process NNCF Quantize by PT

examples/post_training_quantization/openvino/yolov8/main.py

@@ -371,5 +371,24 @@ def main():
     return fp_stats["metrics/mAP50-95(B)"], q_stats["metrics/mAP50-95(B)"], fp_model_perf, quantized_model_perf
 
 
+def check_export_not_strict():
+    model = YOLO(f"{ROOT}/{MODEL_NAME}.pt")
+
+    # Prepare validation dataset and helper
+    validator, data_loader = prepare_validation_new(model, "coco128.yaml")
+
+    batch = next(iter(data_loader))
+    batch = validator.preprocess(batch)
+
+    model.model(batch["img"])
+    ex_model = torch.export.export(model.model, args=(batch["img"],), strict=False)
+    ex_model = capture_pre_autograd_graph(ex_model.module(), args=(batch["img"],))
+
+    fp_stats, total_images, total_objects = validate_fx(ex_model, tqdm(data_loader), validator)
+    print("Floating-point ex strict=False")
+    print_statistics(fp_stats, total_images, total_objects)
+
+
 if __name__ == "__main__":
-    main()
+    check_export_not_strict()
+    # main()

nncf/experimental/torch_fx/nncf_graph_builder.py

@@ -236,6 +236,7 @@ def view_to_reshape(model: torch.fx.GraphModule):
                 continue
             with model.graph.inserting_after(n):
                 reshape = model.graph.create_node("call_function", torch.ops.aten.reshape.default, tuple(n.args), {})
+                reshape.meta = n.meta
 
             for user in list(n.users):
                 user.replace_input_with(n, reshape)
@@ -295,7 +296,7 @@ def create_nncf_graph(model: torch.fx.GraphModule) -> NNCFGraph:
         GraphConverter.separate_conv_and_bias(model)
         GraphConverter.unfold_scaled_dot_product_attention(model)
         GraphConverter.view_to_reshape(model)
-        breakpoint()
+        # breakpoint()
 
         nncf_graph = PTNNCFGraph()
 

nncf/experimental/torch_fx/transformations.py

@@ -105,8 +105,9 @@ def insert_one_qdq(
         dequantize_op = torch.ops.quantized_decomposed.dequantize_per_tensor.default
 
     # Quantized functions accepts only uint8 as an input
-    if target_point.target_type != TargetType.OPERATION_WITH_WEIGHTS and qparams["_dtype_"] == torch.int8:
-        raise RuntimeError("Wrong parameters: activations should always be uint8")
+    # if target_point.target_type != TargetType.OPERATION_WITH_WEIGHTS and qparams["_dtype_"] == torch.int8:
+    #    breakpoint()
+    #    raise RuntimeError("Wrong parameters: activations should always be uint8")
 
     # TODO: map FakeQuantizePramaeters to qparams for quantize/dequantize
     # 2. replace activation_post_process node with quantize and dequantize