TENSOR PREFLIGHT / LOCAL EXECUTION REPORT

Find the failing tensor boundary.

FAIL — inspect before changing code

PyTorch 2.8.0 · Training mode True · Autograd True

Observed result

RuntimeError: mat1 and mat2 shapes cannot be multiplied (2x192 and 64x10)

Deepest unfinished boundary: 1

linear features

Compare the last input dimension with Linear.in_features. After flattening, derive the feature count from the observed tensor and verify axis order before changing the layer.

Changing in_features or reshaping can hide an axis error. Confirm the intended representation rather than selecting dimensions only to make multiplication run.

Module trace

Module	Type	Status	Inputs	Outputs
<root>	Sequential	raised exception	{ "args": [ { "kind": "tensor", "shape": [ 2, 3, 8, 8 ], "dtype": "torch.float32", "device": "cpu", "requires_grad": false } ], "kwargs": {} }	{}
0	Flatten	returned	{ "args": [ { "kind": "tensor", "shape": [ 2, 3, 8, 8 ], "dtype": "torch.float32", "device": "cpu", "requires_grad": false } ], "kwargs": {} }	{ "kind": "tensor", "shape": [ 2, 192 ], "dtype": "torch.float32", "device": "cpu", "requires_grad": false }
1	Linear	exception boundary	{ "args": [ { "kind": "tensor", "shape": [ 2, 192 ], "dtype": "torch.float32", "device": "cpu", "requires_grad": false } ], "kwargs": {} }	{}

Module

Type

Status

Inputs

Outputs

<root>

Sequential

raised exception

{
  "args": [
    {
      "kind": "tensor",
      "shape": [
        2,
        3,
        8,
        8
      ],
      "dtype": "torch.float32",
      "device": "cpu",
      "requires_grad": false
    }
  ],
  "kwargs": {}
}

{}

Flatten

returned

{
  "args": [
    {
      "kind": "tensor",
      "shape": [
        2,
        3,
        8,
        8
      ],
      "dtype": "torch.float32",
      "device": "cpu",
      "requires_grad": false
    }
  ],
  "kwargs": {}
}

{
  "kind": "tensor",
  "shape": [
    2,
    192
  ],
  "dtype": "torch.float32",
  "device": "cpu",
  "requires_grad": false
}

Linear

exception boundary

{
  "args": [
    {
      "kind": "tensor",
      "shape": [
        2,
        192
      ],
      "dtype": "torch.float32",
      "device": "cpu",
      "requires_grad": false
    }
  ],
  "kwargs": {}
}

{}

Output contracts

[]

Parameter and buffer metadata

[
  {
    "module": "<root>",
    "parameters": {},
    "buffers": {}
  },
  {
    "module": "0",
    "parameters": {},
    "buffers": {}
  },
  {
    "module": "1",
    "parameters": {
      "weight": {
        "kind": "tensor",
        "shape": [
          10,
          64
        ],
        "dtype": "torch.float32",
        "device": "cpu",
        "requires_grad": true
      },
      "bias": {
        "kind": "tensor",
        "shape": [
          10
        ],
        "dtype": "torch.float32",
        "device": "cpu",
        "requires_grad": true
      }
    },
    "buffers": {}
  }
]

Scope and sharing

One observed eager forward path; functional operations are not individually traced.
A boundary is not proof of the exact failing operation. Handled child exceptions are recorded but do not imply the overall pass failed.
Execution can mutate model buffers, inputs and random state; training/autograd settings are preserved.
No backward, optimizer, distributed or compiled execution validation. No automatic repair.
Tensor values are omitted, but names, shapes and exception messages may contain private information. Review before sharing.