quark.torch.quantization.graph.optimization.modify_reshape_param
#
Module Contents#
Functions#
- quark.torch.quantization.graph.optimization.modify_reshape_param.modify_reshape_param(m: torch.fx.GraphModule) None #
In some case, reshape param in fx graph is traced by input datashape. For example: reshape.default(conv2d_85, [25, 80, 6400]) # where 25 is batch size we can change: [25, 80, 6400] to [-1, 80, 6400]