Open terfendail opened 7 months ago
If the model is loaded inside of the wrapper
import torch
import torch_mlir
import torchvision
class ModelWrapper(torch.nn.Module):
def __init__(self) -> None:
super().__init__()
self.model = torchvision.models.densenet.densenet161()
self.model.eval()
def forward(self, data):
return self.model(data)[0]
mdl = ModelWrapper()
mdl.eval()
shp = torch_mlir.TensorPlaceholder([-1, 3, 224, 224], torch.float32)
mlir = torch_mlir.compile(mdl, shp, "torch")
the error changes to
Traceback (most recent call last):
File "/pytorch_venv/convert_test.py", line 24, in <module>
mlir = torch_mlir.compile(mdl, shp, "torch")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/pytorch_venv/lib/python3.11/site-packages/torch_mlir/__init__.py", line 458, in compile
run_pipeline_with_repro_report(
File "/pytorch_venv/lib/python3.11/site-packages/torch_mlir/compiler_utils.py", line 73, in run_pipeline_with_repro_report
raise TorchMlirCompilerError(trimmed_message) from None
torch_mlir.compiler_utils.TorchMlirCompilerError: Lowering TorchScript IR -> Torch Backend IR failed with the following diagnostics:
python exception: Failure while executing pass pipeline:
error: "aten::select"("/pytorch_venv/convert_test.py":19:15): unsupported by backend contract: tensor with unknown rank
note: "aten::select"("pytorch_venv/convert_test.py":19:15): see current operation: %844 = "torch.aten.squeeze.dim"(%843, %0) : (!torch.vtensor<[?,1000],f32>, !torch.int) -> !torch.vtensor<*,f32>
note: "aten::select"("/pytorch_venv/convert_test.py":19:15): this is likely due to a missing transfer function in abstract_interp_lib_gen.py
Lowering of DenseNet model fails with error.
Execution of the script cause following error: