ppq使用torch_mlir导出tosa格式的mlir

[zccyman]

def export_mlir(model, dummy_input):

    from torch_mlir import torchscript
    module = torchscript.compile(model, dummy_input, output_type="tosa")
    print("TOSA OutputType\n")
    with open(f"test.mlir", "wt") as f:
        print(module.operation.get_asm(), file=f)

if DEVICE == 'cuda':
    collate_fn = lambda x: x.cuda()
else:
    collate_fn = lambda x: x
   
output = executor.forward(inputs=collate_fn(dataset[0]))
export_mlir(
    executor, 
    collate_fn(dataset[0]),
)        

抛出了如下错误，该怎么解决呢？
Exception has occurred: RuntimeError

Unknown type name 'dict':
File "/home/congcong.zhang/.miniconda3/envs/torch-mlir/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 368
def forward(
self,
inputs: Union[dict, list, torch.Tensor],
~~~~ <--- HERE
output_names:List[str] = None,
hooks: Dict[str, RuntimeHook] = None
File "/home/congcong.zhang/workspace/code/ppq/ppq/samples/test_resnet18.py", line 30, in export_mlir
module = torchscript.compile(model, dummy_input, output_type="tosa")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/congcong.zhang/workspace/code/ppq/ppq/samples/test_resnet18.py", line 246, in
export_mlir(
RuntimeError:
Unknown type name 'dict':
File "/home/congcong.zhang/.miniconda3/envs/torch-mlir/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 368
def forward(
self,
inputs: Union[dict, list, torch.Tensor],
~~~~ <--- HERE
output_names:List[str] = None,
hooks: Dict[str, RuntimeHook] = None

[ZhangZhiPku]

你玩的挺花啊...PPQ有自己的推理引擎，你为啥要导出成mlir呢...