Upload export_medsiglip.py with huggingface_hub

export_medsiglip.py

@@ -212,9 +212,10 @@ def load_trained_model(config: Config):
 
 
 def export_to_onnx(model, config: Config, quantize: bool = True):
-    """Export model to ONNX format"""
+    """Export model to ONNX format using TorchScript tracing"""
     import onnx
-    from onnxruntime.quantization import quantize_dynamic, QuantType
+    import os
+    os.environ["TORCH_ONNX_USE_DYNAMO"] = "0"  # Force legacy exporter
 
     config.output_dir.mkdir(parents=True, exist_ok=True)
 
@@ -222,22 +223,29 @@ def export_to_onnx(model, config: Config, quantize: bool = True):
     wrapper = LaborViewExportWrapper(model)
     wrapper.eval()
 
-    # Create dummy input
+    # Create dummy input on CPU
     dummy_input = torch.randn(1, 3, config.image_size, config.image_size)
 
+    # Move model to CPU for export
+    wrapper = wrapper.cpu()
+
     # Export paths
     onnx_path = config.output_dir / "laborview_medsiglip.onnx"
     onnx_quant_path = config.output_dir / "laborview_medsiglip_int8.onnx"
 
-    print(f"Exporting to ONNX: {onnx_path}")
+    print(f"Exporting to ONNX (legacy exporter): {onnx_path}")
 
-    # Export to ONNX
+    # Trace the model first
+    with torch.no_grad():
+        traced = torch.jit.trace(wrapper, dummy_input)
+
+    # Export traced model to ONNX
     torch.onnx.export(
-        wrapper,
+        traced,
         dummy_input,
         str(onnx_path),
         export_params=True,
-        opset_version=config.opset_version,
+        opset_version=14,  # Use stable opset version
         do_constant_folding=True,
         input_names=['pixel_values'],
         output_names=['seg_probs', 'plane_pred'],
@@ -257,20 +265,27 @@ def export_to_onnx(model, config: Config, quantize: bool = True):
     onnx_size = onnx_path.stat().st_size / (1024 * 1024 * 1024)
     print(f"ONNX model size: {onnx_size:.2f} GB")
 
-    # Quantize to INT8
-    if quantize:
-        print(f"Quantizing to INT8: {onnx_quant_path}")
-        quantize_dynamic(
-            str(onnx_path),
-            str(onnx_quant_path),
-            weight_type=QuantType.QInt8
-        )
+    # Quantize to INT8 (with error handling)
+    onnx_quant_path_result = None
+    if quantize and onnx_size > 0.001:
+        try:
+            from onnxruntime.quantization import quantize_dynamic, QuantType
+            print(f"Quantizing to INT8: {onnx_quant_path}")
+            quantize_dynamic(
+                str(onnx_path),
+                str(onnx_quant_path),
+                weight_type=QuantType.QInt8
+            )
 
-        quant_size = onnx_quant_path.stat().st_size / (1024 * 1024 * 1024)
-        print(f"Quantized model size: {quant_size:.2f} GB")
-        print(f"Size reduction: {(1 - quant_size/onnx_size) * 100:.1f}%")
+            quant_size = onnx_quant_path.stat().st_size / (1024 * 1024 * 1024)
+            print(f"Quantized model size: {quant_size:.2f} GB")
+            print(f"Size reduction: {(1 - quant_size/onnx_size) * 100:.1f}%")
+            onnx_quant_path_result = onnx_quant_path
+        except Exception as e:
+            print(f"Quantization failed: {e}")
+            print("Continuing with FP32 model only")
 
-    return onnx_path, onnx_quant_path if quantize else None
+    return onnx_path, onnx_quant_path_result
 
 
 def export_to_coreml(model, config: Config):