tensor_map.json

{
  "architecture_id": "layout.heron-rtdetr",
  "schema_version": 1,
  "source_config_assertions": [
    {
      "equals": "rt_detr_v2",
      "key": "model_type",
      "path": "config.json"
    },
    {
      "equals": [
        "RTDetrV2ForObjectDetection"
      ],
      "key": "architectures",
      "path": "config.json"
    },
    {
      "equals": 300,
      "key": "num_queries",
      "path": "config.json"
    },
    {
      "equals": 6,
      "key": "decoder_layers",
      "path": "config.json"
    },
    {
      "equals": true,
      "key": "use_focal_loss",
      "path": "config.json"
    },
    {
      "equals": "RTDetrImageProcessor",
      "key": "image_processor_type",
      "path": "preprocessor_config.json"
    }
  ],
  "tensors": [
    {
      "expected_shape": [
        32,
        3,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.0.convolution.weight",
      "target": "model.backbone.model.embedder.embedder.0.convolution.weight"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.0.normalization.bias",
      "target": "model.backbone.model.embedder.embedder.0.normalization.bias"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.0.normalization.running_mean",
      "target": "model.backbone.model.embedder.embedder.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.0.normalization.running_var",
      "target": "model.backbone.model.embedder.embedder.0.normalization.running_var"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.0.normalization.weight",
      "target": "model.backbone.model.embedder.embedder.0.normalization.weight"
    },
    {
      "expected_shape": [
        32,
        32,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.1.convolution.weight",
      "target": "model.backbone.model.embedder.embedder.1.convolution.weight"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.1.normalization.bias",
      "target": "model.backbone.model.embedder.embedder.1.normalization.bias"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.1.normalization.running_mean",
      "target": "model.backbone.model.embedder.embedder.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.1.normalization.running_var",
      "target": "model.backbone.model.embedder.embedder.1.normalization.running_var"
    },
    {
      "expected_shape": [
        32
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.1.normalization.weight",
      "target": "model.backbone.model.embedder.embedder.1.normalization.weight"
    },
    {
      "expected_shape": [
        64,
        32,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.2.convolution.weight",
      "target": "model.backbone.model.embedder.embedder.2.convolution.weight"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.2.normalization.bias",
      "target": "model.backbone.model.embedder.embedder.2.normalization.bias"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.2.normalization.running_mean",
      "target": "model.backbone.model.embedder.embedder.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.2.normalization.running_var",
      "target": "model.backbone.model.embedder.embedder.2.normalization.running_var"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.embedder.embedder.2.normalization.weight",
      "target": "model.backbone.model.embedder.embedder.2.normalization.weight"
    },
    {
      "expected_shape": [
        64,
        64,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        64,
        64,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        64,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        64,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.shortcut.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.shortcut.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.0.shortcut.normalization.weight"
    },
    {
      "expected_shape": [
        64,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        64,
        64,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        64,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.1.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        64,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        64,
        64,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        64
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        64,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.0.layers.2.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        128,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        128,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.0.shortcut.1.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        128,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        128,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.1.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        128,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        128,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.2.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        128,
        128,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        128
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        128,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.1.layers.3.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        1024,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        1024,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.convolution.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.bias"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.running_var"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.0.shortcut.1.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        1024,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.1.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        1024,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.2.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        1024,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.3.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        1024,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.4.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        1024,
        256,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.2.layers.5.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        512,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        2048,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        2048,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.convolution.weight"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.bias"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.running_var"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.0.shortcut.1.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        2048,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        512,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        2048,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.1.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        2048,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.0.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.0.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.0.normalization.weight"
    },
    {
      "expected_shape": [
        512,
        512,
        3,
        3
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.1.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.1.convolution.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.bias"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.running_mean"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.running_var"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.1.normalization.weight"
    },
    {
      "expected_shape": [
        2048,
        512,
        1,
        1
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.2.convolution.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.2.convolution.weight"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.bias",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.bias"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.running_mean",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.running_mean"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.running_var",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.running_var"
    },
    {
      "expected_shape": [
        2048
      ],
      "notes": "RT-DETR ResNet backbone tensor reused without transpose",
      "source": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.weight",
      "target": "model.backbone.model.encoder.stages.3.layers.2.layer.2.normalization.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.0.layers.0.bias",
      "target": "model.decoder.bbox_embed.0.layers.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.0.layers.0.weight",
      "target": "model.decoder.bbox_embed.0.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.0.layers.1.bias",
      "target": "model.decoder.bbox_embed.0.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.0.layers.1.weight",
      "target": "model.decoder.bbox_embed.0.layers.1.weight"
    },
    {
      "expected_shape": [
        4
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.0.layers.2.bias",
      "target": "model.decoder.bbox_embed.0.layers.2.bias"
    },
    {
      "expected_shape": [
        4,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.0.layers.2.weight",
      "target": "model.decoder.bbox_embed.0.layers.2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.1.layers.0.bias",
      "target": "model.decoder.bbox_embed.1.layers.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.1.layers.0.weight",
      "target": "model.decoder.bbox_embed.1.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.1.layers.1.bias",
      "target": "model.decoder.bbox_embed.1.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.1.layers.1.weight",
      "target": "model.decoder.bbox_embed.1.layers.1.weight"
    },
    {
      "expected_shape": [
        4
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.1.layers.2.bias",
      "target": "model.decoder.bbox_embed.1.layers.2.bias"
    },
    {
      "expected_shape": [
        4,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.1.layers.2.weight",
      "target": "model.decoder.bbox_embed.1.layers.2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.2.layers.0.bias",
      "target": "model.decoder.bbox_embed.2.layers.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.2.layers.0.weight",
      "target": "model.decoder.bbox_embed.2.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.2.layers.1.bias",
      "target": "model.decoder.bbox_embed.2.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.2.layers.1.weight",
      "target": "model.decoder.bbox_embed.2.layers.1.weight"
    },
    {
      "expected_shape": [
        4
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.2.layers.2.bias",
      "target": "model.decoder.bbox_embed.2.layers.2.bias"
    },
    {
      "expected_shape": [
        4,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.2.layers.2.weight",
      "target": "model.decoder.bbox_embed.2.layers.2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.3.layers.0.bias",
      "target": "model.decoder.bbox_embed.3.layers.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.3.layers.0.weight",
      "target": "model.decoder.bbox_embed.3.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.3.layers.1.bias",
      "target": "model.decoder.bbox_embed.3.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.3.layers.1.weight",
      "target": "model.decoder.bbox_embed.3.layers.1.weight"
    },
    {
      "expected_shape": [
        4
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.3.layers.2.bias",
      "target": "model.decoder.bbox_embed.3.layers.2.bias"
    },
    {
      "expected_shape": [
        4,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.3.layers.2.weight",
      "target": "model.decoder.bbox_embed.3.layers.2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.4.layers.0.bias",
      "target": "model.decoder.bbox_embed.4.layers.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.4.layers.0.weight",
      "target": "model.decoder.bbox_embed.4.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.4.layers.1.bias",
      "target": "model.decoder.bbox_embed.4.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.4.layers.1.weight",
      "target": "model.decoder.bbox_embed.4.layers.1.weight"
    },
    {
      "expected_shape": [
        4
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.4.layers.2.bias",
      "target": "model.decoder.bbox_embed.4.layers.2.bias"
    },
    {
      "expected_shape": [
        4,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.4.layers.2.weight",
      "target": "model.decoder.bbox_embed.4.layers.2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.5.layers.0.bias",
      "target": "model.decoder.bbox_embed.5.layers.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.5.layers.0.weight",
      "target": "model.decoder.bbox_embed.5.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.5.layers.1.bias",
      "target": "model.decoder.bbox_embed.5.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.5.layers.1.weight",
      "target": "model.decoder.bbox_embed.5.layers.1.weight"
    },
    {
      "expected_shape": [
        4
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.5.layers.2.bias",
      "target": "model.decoder.bbox_embed.5.layers.2.bias"
    },
    {
      "expected_shape": [
        4,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.bbox_embed.5.layers.2.weight",
      "target": "model.decoder.bbox_embed.5.layers.2.weight"
    },
    {
      "expected_shape": [
        17
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.0.bias",
      "target": "model.decoder.class_embed.0.bias"
    },
    {
      "expected_shape": [
        17,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.0.weight",
      "target": "model.decoder.class_embed.0.weight"
    },
    {
      "expected_shape": [
        17
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.1.bias",
      "target": "model.decoder.class_embed.1.bias"
    },
    {
      "expected_shape": [
        17,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.1.weight",
      "target": "model.decoder.class_embed.1.weight"
    },
    {
      "expected_shape": [
        17
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.2.bias",
      "target": "model.decoder.class_embed.2.bias"
    },
    {
      "expected_shape": [
        17,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.2.weight",
      "target": "model.decoder.class_embed.2.weight"
    },
    {
      "expected_shape": [
        17
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.3.bias",
      "target": "model.decoder.class_embed.3.bias"
    },
    {
      "expected_shape": [
        17,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.3.weight",
      "target": "model.decoder.class_embed.3.weight"
    },
    {
      "expected_shape": [
        17
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.4.bias",
      "target": "model.decoder.class_embed.4.bias"
    },
    {
      "expected_shape": [
        17,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.4.weight",
      "target": "model.decoder.class_embed.4.weight"
    },
    {
      "expected_shape": [
        17
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.5.bias",
      "target": "model.decoder.class_embed.5.bias"
    },
    {
      "expected_shape": [
        17,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.class_embed.5.weight",
      "target": "model.decoder.class_embed.5.weight"
    },
    {
      "expected_shape": [
        96
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.attention_weights.bias",
      "target": "model.decoder.layers.0.encoder_attn.attention_weights.bias"
    },
    {
      "expected_shape": [
        96,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.attention_weights.weight",
      "target": "model.decoder.layers.0.encoder_attn.attention_weights.weight"
    },
    {
      "expected_shape": [
        12
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.n_points_scale",
      "target": "model.decoder.layers.0.encoder_attn.n_points_scale"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.output_proj.bias",
      "target": "model.decoder.layers.0.encoder_attn.output_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.output_proj.weight",
      "target": "model.decoder.layers.0.encoder_attn.output_proj.weight"
    },
    {
      "expected_shape": [
        192
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.sampling_offsets.bias",
      "target": "model.decoder.layers.0.encoder_attn.sampling_offsets.bias"
    },
    {
      "expected_shape": [
        192,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.sampling_offsets.weight",
      "target": "model.decoder.layers.0.encoder_attn.sampling_offsets.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.value_proj.bias",
      "target": "model.decoder.layers.0.encoder_attn.value_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn.value_proj.weight",
      "target": "model.decoder.layers.0.encoder_attn.value_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn_layer_norm.bias",
      "target": "model.decoder.layers.0.encoder_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.encoder_attn_layer_norm.weight",
      "target": "model.decoder.layers.0.encoder_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.fc1.bias",
      "target": "model.decoder.layers.0.fc1.bias"
    },
    {
      "expected_shape": [
        1024,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.fc1.weight",
      "target": "model.decoder.layers.0.fc1.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.fc2.bias",
      "target": "model.decoder.layers.0.fc2.bias"
    },
    {
      "expected_shape": [
        256,
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.fc2.weight",
      "target": "model.decoder.layers.0.fc2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.final_layer_norm.bias",
      "target": "model.decoder.layers.0.final_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.final_layer_norm.weight",
      "target": "model.decoder.layers.0.final_layer_norm.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.k_proj.bias",
      "target": "model.decoder.layers.0.self_attn.k_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.k_proj.weight",
      "target": "model.decoder.layers.0.self_attn.k_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.out_proj.bias",
      "target": "model.decoder.layers.0.self_attn.out_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.out_proj.weight",
      "target": "model.decoder.layers.0.self_attn.out_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.q_proj.bias",
      "target": "model.decoder.layers.0.self_attn.q_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.q_proj.weight",
      "target": "model.decoder.layers.0.self_attn.q_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.v_proj.bias",
      "target": "model.decoder.layers.0.self_attn.v_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn.v_proj.weight",
      "target": "model.decoder.layers.0.self_attn.v_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn_layer_norm.bias",
      "target": "model.decoder.layers.0.self_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.0.self_attn_layer_norm.weight",
      "target": "model.decoder.layers.0.self_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        96
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.attention_weights.bias",
      "target": "model.decoder.layers.1.encoder_attn.attention_weights.bias"
    },
    {
      "expected_shape": [
        96,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.attention_weights.weight",
      "target": "model.decoder.layers.1.encoder_attn.attention_weights.weight"
    },
    {
      "expected_shape": [
        12
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.n_points_scale",
      "target": "model.decoder.layers.1.encoder_attn.n_points_scale"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.output_proj.bias",
      "target": "model.decoder.layers.1.encoder_attn.output_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.output_proj.weight",
      "target": "model.decoder.layers.1.encoder_attn.output_proj.weight"
    },
    {
      "expected_shape": [
        192
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.sampling_offsets.bias",
      "target": "model.decoder.layers.1.encoder_attn.sampling_offsets.bias"
    },
    {
      "expected_shape": [
        192,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.sampling_offsets.weight",
      "target": "model.decoder.layers.1.encoder_attn.sampling_offsets.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.value_proj.bias",
      "target": "model.decoder.layers.1.encoder_attn.value_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn.value_proj.weight",
      "target": "model.decoder.layers.1.encoder_attn.value_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn_layer_norm.bias",
      "target": "model.decoder.layers.1.encoder_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.encoder_attn_layer_norm.weight",
      "target": "model.decoder.layers.1.encoder_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.fc1.bias",
      "target": "model.decoder.layers.1.fc1.bias"
    },
    {
      "expected_shape": [
        1024,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.fc1.weight",
      "target": "model.decoder.layers.1.fc1.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.fc2.bias",
      "target": "model.decoder.layers.1.fc2.bias"
    },
    {
      "expected_shape": [
        256,
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.fc2.weight",
      "target": "model.decoder.layers.1.fc2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.final_layer_norm.bias",
      "target": "model.decoder.layers.1.final_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.final_layer_norm.weight",
      "target": "model.decoder.layers.1.final_layer_norm.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.k_proj.bias",
      "target": "model.decoder.layers.1.self_attn.k_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.k_proj.weight",
      "target": "model.decoder.layers.1.self_attn.k_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.out_proj.bias",
      "target": "model.decoder.layers.1.self_attn.out_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.out_proj.weight",
      "target": "model.decoder.layers.1.self_attn.out_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.q_proj.bias",
      "target": "model.decoder.layers.1.self_attn.q_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.q_proj.weight",
      "target": "model.decoder.layers.1.self_attn.q_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.v_proj.bias",
      "target": "model.decoder.layers.1.self_attn.v_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn.v_proj.weight",
      "target": "model.decoder.layers.1.self_attn.v_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn_layer_norm.bias",
      "target": "model.decoder.layers.1.self_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.1.self_attn_layer_norm.weight",
      "target": "model.decoder.layers.1.self_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        96
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.attention_weights.bias",
      "target": "model.decoder.layers.2.encoder_attn.attention_weights.bias"
    },
    {
      "expected_shape": [
        96,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.attention_weights.weight",
      "target": "model.decoder.layers.2.encoder_attn.attention_weights.weight"
    },
    {
      "expected_shape": [
        12
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.n_points_scale",
      "target": "model.decoder.layers.2.encoder_attn.n_points_scale"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.output_proj.bias",
      "target": "model.decoder.layers.2.encoder_attn.output_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.output_proj.weight",
      "target": "model.decoder.layers.2.encoder_attn.output_proj.weight"
    },
    {
      "expected_shape": [
        192
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.sampling_offsets.bias",
      "target": "model.decoder.layers.2.encoder_attn.sampling_offsets.bias"
    },
    {
      "expected_shape": [
        192,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.sampling_offsets.weight",
      "target": "model.decoder.layers.2.encoder_attn.sampling_offsets.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.value_proj.bias",
      "target": "model.decoder.layers.2.encoder_attn.value_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn.value_proj.weight",
      "target": "model.decoder.layers.2.encoder_attn.value_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn_layer_norm.bias",
      "target": "model.decoder.layers.2.encoder_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.encoder_attn_layer_norm.weight",
      "target": "model.decoder.layers.2.encoder_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.fc1.bias",
      "target": "model.decoder.layers.2.fc1.bias"
    },
    {
      "expected_shape": [
        1024,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.fc1.weight",
      "target": "model.decoder.layers.2.fc1.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.fc2.bias",
      "target": "model.decoder.layers.2.fc2.bias"
    },
    {
      "expected_shape": [
        256,
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.fc2.weight",
      "target": "model.decoder.layers.2.fc2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.final_layer_norm.bias",
      "target": "model.decoder.layers.2.final_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.final_layer_norm.weight",
      "target": "model.decoder.layers.2.final_layer_norm.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.k_proj.bias",
      "target": "model.decoder.layers.2.self_attn.k_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.k_proj.weight",
      "target": "model.decoder.layers.2.self_attn.k_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.out_proj.bias",
      "target": "model.decoder.layers.2.self_attn.out_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.out_proj.weight",
      "target": "model.decoder.layers.2.self_attn.out_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.q_proj.bias",
      "target": "model.decoder.layers.2.self_attn.q_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.q_proj.weight",
      "target": "model.decoder.layers.2.self_attn.q_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.v_proj.bias",
      "target": "model.decoder.layers.2.self_attn.v_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn.v_proj.weight",
      "target": "model.decoder.layers.2.self_attn.v_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn_layer_norm.bias",
      "target": "model.decoder.layers.2.self_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.2.self_attn_layer_norm.weight",
      "target": "model.decoder.layers.2.self_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        96
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.attention_weights.bias",
      "target": "model.decoder.layers.3.encoder_attn.attention_weights.bias"
    },
    {
      "expected_shape": [
        96,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.attention_weights.weight",
      "target": "model.decoder.layers.3.encoder_attn.attention_weights.weight"
    },
    {
      "expected_shape": [
        12
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.n_points_scale",
      "target": "model.decoder.layers.3.encoder_attn.n_points_scale"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.output_proj.bias",
      "target": "model.decoder.layers.3.encoder_attn.output_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.output_proj.weight",
      "target": "model.decoder.layers.3.encoder_attn.output_proj.weight"
    },
    {
      "expected_shape": [
        192
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.sampling_offsets.bias",
      "target": "model.decoder.layers.3.encoder_attn.sampling_offsets.bias"
    },
    {
      "expected_shape": [
        192,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.sampling_offsets.weight",
      "target": "model.decoder.layers.3.encoder_attn.sampling_offsets.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.value_proj.bias",
      "target": "model.decoder.layers.3.encoder_attn.value_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn.value_proj.weight",
      "target": "model.decoder.layers.3.encoder_attn.value_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn_layer_norm.bias",
      "target": "model.decoder.layers.3.encoder_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.encoder_attn_layer_norm.weight",
      "target": "model.decoder.layers.3.encoder_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.fc1.bias",
      "target": "model.decoder.layers.3.fc1.bias"
    },
    {
      "expected_shape": [
        1024,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.fc1.weight",
      "target": "model.decoder.layers.3.fc1.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.fc2.bias",
      "target": "model.decoder.layers.3.fc2.bias"
    },
    {
      "expected_shape": [
        256,
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.fc2.weight",
      "target": "model.decoder.layers.3.fc2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.final_layer_norm.bias",
      "target": "model.decoder.layers.3.final_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.final_layer_norm.weight",
      "target": "model.decoder.layers.3.final_layer_norm.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.k_proj.bias",
      "target": "model.decoder.layers.3.self_attn.k_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.k_proj.weight",
      "target": "model.decoder.layers.3.self_attn.k_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.out_proj.bias",
      "target": "model.decoder.layers.3.self_attn.out_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.out_proj.weight",
      "target": "model.decoder.layers.3.self_attn.out_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.q_proj.bias",
      "target": "model.decoder.layers.3.self_attn.q_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.q_proj.weight",
      "target": "model.decoder.layers.3.self_attn.q_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.v_proj.bias",
      "target": "model.decoder.layers.3.self_attn.v_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn.v_proj.weight",
      "target": "model.decoder.layers.3.self_attn.v_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn_layer_norm.bias",
      "target": "model.decoder.layers.3.self_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.3.self_attn_layer_norm.weight",
      "target": "model.decoder.layers.3.self_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        96
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.attention_weights.bias",
      "target": "model.decoder.layers.4.encoder_attn.attention_weights.bias"
    },
    {
      "expected_shape": [
        96,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.attention_weights.weight",
      "target": "model.decoder.layers.4.encoder_attn.attention_weights.weight"
    },
    {
      "expected_shape": [
        12
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.n_points_scale",
      "target": "model.decoder.layers.4.encoder_attn.n_points_scale"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.output_proj.bias",
      "target": "model.decoder.layers.4.encoder_attn.output_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.output_proj.weight",
      "target": "model.decoder.layers.4.encoder_attn.output_proj.weight"
    },
    {
      "expected_shape": [
        192
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.sampling_offsets.bias",
      "target": "model.decoder.layers.4.encoder_attn.sampling_offsets.bias"
    },
    {
      "expected_shape": [
        192,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.sampling_offsets.weight",
      "target": "model.decoder.layers.4.encoder_attn.sampling_offsets.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.value_proj.bias",
      "target": "model.decoder.layers.4.encoder_attn.value_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn.value_proj.weight",
      "target": "model.decoder.layers.4.encoder_attn.value_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn_layer_norm.bias",
      "target": "model.decoder.layers.4.encoder_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.encoder_attn_layer_norm.weight",
      "target": "model.decoder.layers.4.encoder_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.fc1.bias",
      "target": "model.decoder.layers.4.fc1.bias"
    },
    {
      "expected_shape": [
        1024,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.fc1.weight",
      "target": "model.decoder.layers.4.fc1.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.fc2.bias",
      "target": "model.decoder.layers.4.fc2.bias"
    },
    {
      "expected_shape": [
        256,
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.fc2.weight",
      "target": "model.decoder.layers.4.fc2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.final_layer_norm.bias",
      "target": "model.decoder.layers.4.final_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.final_layer_norm.weight",
      "target": "model.decoder.layers.4.final_layer_norm.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.k_proj.bias",
      "target": "model.decoder.layers.4.self_attn.k_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.k_proj.weight",
      "target": "model.decoder.layers.4.self_attn.k_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.out_proj.bias",
      "target": "model.decoder.layers.4.self_attn.out_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.out_proj.weight",
      "target": "model.decoder.layers.4.self_attn.out_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.q_proj.bias",
      "target": "model.decoder.layers.4.self_attn.q_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.q_proj.weight",
      "target": "model.decoder.layers.4.self_attn.q_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.v_proj.bias",
      "target": "model.decoder.layers.4.self_attn.v_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn.v_proj.weight",
      "target": "model.decoder.layers.4.self_attn.v_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn_layer_norm.bias",
      "target": "model.decoder.layers.4.self_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.4.self_attn_layer_norm.weight",
      "target": "model.decoder.layers.4.self_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        96
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.attention_weights.bias",
      "target": "model.decoder.layers.5.encoder_attn.attention_weights.bias"
    },
    {
      "expected_shape": [
        96,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.attention_weights.weight",
      "target": "model.decoder.layers.5.encoder_attn.attention_weights.weight"
    },
    {
      "expected_shape": [
        12
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.n_points_scale",
      "target": "model.decoder.layers.5.encoder_attn.n_points_scale"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.output_proj.bias",
      "target": "model.decoder.layers.5.encoder_attn.output_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.output_proj.weight",
      "target": "model.decoder.layers.5.encoder_attn.output_proj.weight"
    },
    {
      "expected_shape": [
        192
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.sampling_offsets.bias",
      "target": "model.decoder.layers.5.encoder_attn.sampling_offsets.bias"
    },
    {
      "expected_shape": [
        192,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.sampling_offsets.weight",
      "target": "model.decoder.layers.5.encoder_attn.sampling_offsets.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.value_proj.bias",
      "target": "model.decoder.layers.5.encoder_attn.value_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn.value_proj.weight",
      "target": "model.decoder.layers.5.encoder_attn.value_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn_layer_norm.bias",
      "target": "model.decoder.layers.5.encoder_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.encoder_attn_layer_norm.weight",
      "target": "model.decoder.layers.5.encoder_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.fc1.bias",
      "target": "model.decoder.layers.5.fc1.bias"
    },
    {
      "expected_shape": [
        1024,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.fc1.weight",
      "target": "model.decoder.layers.5.fc1.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.fc2.bias",
      "target": "model.decoder.layers.5.fc2.bias"
    },
    {
      "expected_shape": [
        256,
        1024
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.fc2.weight",
      "target": "model.decoder.layers.5.fc2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.final_layer_norm.bias",
      "target": "model.decoder.layers.5.final_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.final_layer_norm.weight",
      "target": "model.decoder.layers.5.final_layer_norm.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.k_proj.bias",
      "target": "model.decoder.layers.5.self_attn.k_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.k_proj.weight",
      "target": "model.decoder.layers.5.self_attn.k_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.out_proj.bias",
      "target": "model.decoder.layers.5.self_attn.out_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.out_proj.weight",
      "target": "model.decoder.layers.5.self_attn.out_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.q_proj.bias",
      "target": "model.decoder.layers.5.self_attn.q_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.q_proj.weight",
      "target": "model.decoder.layers.5.self_attn.q_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.v_proj.bias",
      "target": "model.decoder.layers.5.self_attn.v_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn.v_proj.weight",
      "target": "model.decoder.layers.5.self_attn.v_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn_layer_norm.bias",
      "target": "model.decoder.layers.5.self_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.layers.5.self_attn_layer_norm.weight",
      "target": "model.decoder.layers.5.self_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        512
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.query_pos_head.layers.0.bias",
      "target": "model.decoder.query_pos_head.layers.0.bias"
    },
    {
      "expected_shape": [
        512,
        4
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.query_pos_head.layers.0.weight",
      "target": "model.decoder.query_pos_head.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.query_pos_head.layers.1.bias",
      "target": "model.decoder.query_pos_head.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        512
      ],
      "notes": "RT-DETR decoder tensor reused without transpose",
      "source": "model.decoder.query_pos_head.layers.1.weight",
      "target": "model.decoder.query_pos_head.layers.1.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.0.0.weight",
      "target": "model.decoder_input_proj.0.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.0.1.bias",
      "target": "model.decoder_input_proj.0.1.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.decoder_input_proj.0.1.num_batches_tracked",
      "target": "model.decoder_input_proj.0.1.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.0.1.running_mean",
      "target": "model.decoder_input_proj.0.1.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.0.1.running_var",
      "target": "model.decoder_input_proj.0.1.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.0.1.weight",
      "target": "model.decoder_input_proj.0.1.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.1.0.weight",
      "target": "model.decoder_input_proj.1.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.1.1.bias",
      "target": "model.decoder_input_proj.1.1.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.decoder_input_proj.1.1.num_batches_tracked",
      "target": "model.decoder_input_proj.1.1.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.1.1.running_mean",
      "target": "model.decoder_input_proj.1.1.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.1.1.running_var",
      "target": "model.decoder_input_proj.1.1.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.1.1.weight",
      "target": "model.decoder_input_proj.1.1.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.2.0.weight",
      "target": "model.decoder_input_proj.2.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.2.1.bias",
      "target": "model.decoder_input_proj.2.1.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.decoder_input_proj.2.1.num_batches_tracked",
      "target": "model.decoder_input_proj.2.1.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.2.1.running_mean",
      "target": "model.decoder_input_proj.2.1.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.2.1.running_var",
      "target": "model.decoder_input_proj.2.1.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR decoder input projection tensor reused without transpose",
      "source": "model.decoder_input_proj.2.1.weight",
      "target": "model.decoder_input_proj.2.1.weight"
    },
    {
      "expected_shape": [
        18,
        256
      ],
      "notes": "RT-DETR denoising class embedding reused without transpose",
      "source": "model.denoising_class_embed.weight",
      "target": "model.denoising_class_embed.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder bbox head tensor reused without transpose",
      "source": "model.enc_bbox_head.layers.0.bias",
      "target": "model.enc_bbox_head.layers.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR encoder bbox head tensor reused without transpose",
      "source": "model.enc_bbox_head.layers.0.weight",
      "target": "model.enc_bbox_head.layers.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder bbox head tensor reused without transpose",
      "source": "model.enc_bbox_head.layers.1.bias",
      "target": "model.enc_bbox_head.layers.1.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR encoder bbox head tensor reused without transpose",
      "source": "model.enc_bbox_head.layers.1.weight",
      "target": "model.enc_bbox_head.layers.1.weight"
    },
    {
      "expected_shape": [
        4
      ],
      "notes": "RT-DETR encoder bbox head tensor reused without transpose",
      "source": "model.enc_bbox_head.layers.2.bias",
      "target": "model.enc_bbox_head.layers.2.bias"
    },
    {
      "expected_shape": [
        4,
        256
      ],
      "notes": "RT-DETR encoder bbox head tensor reused without transpose",
      "source": "model.enc_bbox_head.layers.2.weight",
      "target": "model.enc_bbox_head.layers.2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder output projection tensor reused without transpose",
      "source": "model.enc_output.0.bias",
      "target": "model.enc_output.0.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR encoder output projection tensor reused without transpose",
      "source": "model.enc_output.0.weight",
      "target": "model.enc_output.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder output projection tensor reused without transpose",
      "source": "model.enc_output.1.bias",
      "target": "model.enc_output.1.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder output projection tensor reused without transpose",
      "source": "model.enc_output.1.weight",
      "target": "model.enc_output.1.weight"
    },
    {
      "expected_shape": [
        17
      ],
      "notes": "RT-DETR encoder score head tensor reused without transpose",
      "source": "model.enc_score_head.bias",
      "target": "model.enc_score_head.bias"
    },
    {
      "expected_shape": [
        17,
        256
      ],
      "notes": "RT-DETR encoder score head tensor reused without transpose",
      "source": "model.enc_score_head.weight",
      "target": "model.enc_score_head.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.0.conv.weight",
      "target": "model.encoder.downsample_convs.0.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.0.norm.bias",
      "target": "model.encoder.downsample_convs.0.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.downsample_convs.0.norm.num_batches_tracked",
      "target": "model.encoder.downsample_convs.0.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.0.norm.running_mean",
      "target": "model.encoder.downsample_convs.0.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.0.norm.running_var",
      "target": "model.encoder.downsample_convs.0.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.0.norm.weight",
      "target": "model.encoder.downsample_convs.0.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.1.conv.weight",
      "target": "model.encoder.downsample_convs.1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.1.norm.bias",
      "target": "model.encoder.downsample_convs.1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.downsample_convs.1.norm.num_batches_tracked",
      "target": "model.encoder.downsample_convs.1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.1.norm.running_mean",
      "target": "model.encoder.downsample_convs.1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.1.norm.running_var",
      "target": "model.encoder.downsample_convs.1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.downsample_convs.1.norm.weight",
      "target": "model.encoder.downsample_convs.1.norm.weight"
    },
    {
      "expected_shape": [
        1024
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.fc1.bias",
      "target": "model.encoder.encoder.0.layers.0.fc1.bias"
    },
    {
      "expected_shape": [
        1024,
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.fc1.weight",
      "target": "model.encoder.encoder.0.layers.0.fc1.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.fc2.bias",
      "target": "model.encoder.encoder.0.layers.0.fc2.bias"
    },
    {
      "expected_shape": [
        256,
        1024
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.fc2.weight",
      "target": "model.encoder.encoder.0.layers.0.fc2.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.final_layer_norm.bias",
      "target": "model.encoder.encoder.0.layers.0.final_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.final_layer_norm.weight",
      "target": "model.encoder.encoder.0.layers.0.final_layer_norm.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.k_proj.bias",
      "target": "model.encoder.encoder.0.layers.0.self_attn.k_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.k_proj.weight",
      "target": "model.encoder.encoder.0.layers.0.self_attn.k_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.out_proj.bias",
      "target": "model.encoder.encoder.0.layers.0.self_attn.out_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.out_proj.weight",
      "target": "model.encoder.encoder.0.layers.0.self_attn.out_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.q_proj.bias",
      "target": "model.encoder.encoder.0.layers.0.self_attn.q_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.q_proj.weight",
      "target": "model.encoder.encoder.0.layers.0.self_attn.q_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.v_proj.bias",
      "target": "model.encoder.encoder.0.layers.0.self_attn.v_proj.bias"
    },
    {
      "expected_shape": [
        256,
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn.v_proj.weight",
      "target": "model.encoder.encoder.0.layers.0.self_attn.v_proj.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn_layer_norm.bias",
      "target": "model.encoder.encoder.0.layers.0.self_attn_layer_norm.bias"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.encoder.0.layers.0.self_attn_layer_norm.weight",
      "target": "model.encoder.encoder.0.layers.0.self_attn_layer_norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.0.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.1.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.0.bottlenecks.2.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.0.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.0.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.0.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.0.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.0.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.0.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.0.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.0.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.0.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.0.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.0.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.0.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.1.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.1.bottlenecks.2.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv1.conv.weight",
      "target": "model.encoder.fpn_blocks.1.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv1.norm.bias",
      "target": "model.encoder.fpn_blocks.1.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.conv1.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv1.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv1.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv1.norm.weight",
      "target": "model.encoder.fpn_blocks.1.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv2.conv.weight",
      "target": "model.encoder.fpn_blocks.1.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv2.norm.bias",
      "target": "model.encoder.fpn_blocks.1.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.fpn_blocks.1.conv2.norm.num_batches_tracked",
      "target": "model.encoder.fpn_blocks.1.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv2.norm.running_mean",
      "target": "model.encoder.fpn_blocks.1.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv2.norm.running_var",
      "target": "model.encoder.fpn_blocks.1.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.fpn_blocks.1.conv2.norm.weight",
      "target": "model.encoder.fpn_blocks.1.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.0.conv.weight",
      "target": "model.encoder.lateral_convs.0.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.0.norm.bias",
      "target": "model.encoder.lateral_convs.0.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.lateral_convs.0.norm.num_batches_tracked",
      "target": "model.encoder.lateral_convs.0.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.0.norm.running_mean",
      "target": "model.encoder.lateral_convs.0.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.0.norm.running_var",
      "target": "model.encoder.lateral_convs.0.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.0.norm.weight",
      "target": "model.encoder.lateral_convs.0.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.1.conv.weight",
      "target": "model.encoder.lateral_convs.1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.1.norm.bias",
      "target": "model.encoder.lateral_convs.1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.lateral_convs.1.norm.num_batches_tracked",
      "target": "model.encoder.lateral_convs.1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.1.norm.running_mean",
      "target": "model.encoder.lateral_convs.1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.1.norm.running_var",
      "target": "model.encoder.lateral_convs.1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.lateral_convs.1.norm.weight",
      "target": "model.encoder.lateral_convs.1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.0.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.1.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.0.bottlenecks.2.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.0.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.0.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.0.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.0.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.0.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.0.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.0.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.0.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.0.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.0.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.0.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.0.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.0.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.1.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        3,
        3
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        256,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.1.bottlenecks.2.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv1.conv.weight",
      "target": "model.encoder.pan_blocks.1.conv1.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv1.norm.bias",
      "target": "model.encoder.pan_blocks.1.conv1.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.conv1.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.conv1.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv1.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.conv1.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv1.norm.running_var",
      "target": "model.encoder.pan_blocks.1.conv1.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv1.norm.weight",
      "target": "model.encoder.pan_blocks.1.conv1.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv2.conv.weight",
      "target": "model.encoder.pan_blocks.1.conv2.conv.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv2.norm.bias",
      "target": "model.encoder.pan_blocks.1.conv2.norm.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder.pan_blocks.1.conv2.norm.num_batches_tracked",
      "target": "model.encoder.pan_blocks.1.conv2.norm.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv2.norm.running_mean",
      "target": "model.encoder.pan_blocks.1.conv2.norm.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv2.norm.running_var",
      "target": "model.encoder.pan_blocks.1.conv2.norm.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR hybrid encoder, FPN, or PAN tensor reused without transpose",
      "source": "model.encoder.pan_blocks.1.conv2.norm.weight",
      "target": "model.encoder.pan_blocks.1.conv2.norm.weight"
    },
    {
      "expected_shape": [
        256,
        512,
        1,
        1
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.0.0.weight",
      "target": "model.encoder_input_proj.0.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.0.1.bias",
      "target": "model.encoder_input_proj.0.1.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder_input_proj.0.1.num_batches_tracked",
      "target": "model.encoder_input_proj.0.1.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.0.1.running_mean",
      "target": "model.encoder_input_proj.0.1.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.0.1.running_var",
      "target": "model.encoder_input_proj.0.1.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.0.1.weight",
      "target": "model.encoder_input_proj.0.1.weight"
    },
    {
      "expected_shape": [
        256,
        1024,
        1,
        1
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.1.0.weight",
      "target": "model.encoder_input_proj.1.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.1.1.bias",
      "target": "model.encoder_input_proj.1.1.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder_input_proj.1.1.num_batches_tracked",
      "target": "model.encoder_input_proj.1.1.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.1.1.running_mean",
      "target": "model.encoder_input_proj.1.1.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.1.1.running_var",
      "target": "model.encoder_input_proj.1.1.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.1.1.weight",
      "target": "model.encoder_input_proj.1.1.weight"
    },
    {
      "expected_shape": [
        256,
        2048,
        1,
        1
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.2.0.weight",
      "target": "model.encoder_input_proj.2.0.weight"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.2.1.bias",
      "target": "model.encoder_input_proj.2.1.bias"
    },
    {
      "expected_shape": [
        "scalar"
      ],
      "notes": "batch norm training counter copied from source; native inference may ignore it",
      "source": "model.encoder_input_proj.2.1.num_batches_tracked",
      "target": "model.encoder_input_proj.2.1.num_batches_tracked"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.2.1.running_mean",
      "target": "model.encoder_input_proj.2.1.running_mean"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.2.1.running_var",
      "target": "model.encoder_input_proj.2.1.running_var"
    },
    {
      "expected_shape": [
        256
      ],
      "notes": "RT-DETR encoder input projection tensor reused without transpose",
      "source": "model.encoder_input_proj.2.1.weight",
      "target": "model.encoder_input_proj.2.1.weight"
    }
  ]
}