To convert a quantized model to an XMODEL, vai_q_pytorch offers a decorator that
allows you to register an operation or a group of operations as a custom operation that
XIR does not
recognize:
# Decorator API
def register_custom_op(op_type: str, attrs_list: Optional[List[str]] = None):
"""The decorator is used to register the function as a custom operation.
Args:
op_type(str) - the operator type registered into quantizer.
The type should not conflict with pytorch_nndct
attrs_list(Optional[List[str]], optional) -
the name list of attributes that define operation flavor.
For example, Convolution operation has such attributes as padding, dilation, stride and groups.
The order of names in attrs_list should be consistent with that of the arguments list.
Default: None
"""
Perform the following steps:
- Aggregate some operations as a function. The first argument name of this function should be ctx, with the same meaning as in torch.autograd.Function
- Decorate this function with the decorator described previously.
from pytorch_nndct.utils import register_custom_op
@register_custom_op(op_type="MyOp", attrs_list=["scale_1", "scale_2"])
def custom_op(ctx, x: torch.Tensor, y:torch.Tensor, scale_1:float, scale_2:float) -> torch.Tensor:
return scale_1 * x + scale_2 * y
class MyModule(torch.nn.Module):
def __init__(self):
...
def forward(self, x, y):
return custom_op(x, y, scale_1=2.0, scale_2=1.0)
Limitations: - Loop operation is not allowed in a custom operation.
- A custom operation's number of return values can only be one.