def fuse_modules(model, modules_to_fuse, inplace=False, fuser_func=fuse_known_modules, fuse_custom_config_dict=None):

def fuse_known_modules(mod_list, is_qat, additional_fuser_method_mapping=None):
    r"""Returns a list of modules that fuses the operations specified
     in the input module list.
    Fuses only the following sequence of modules:
    conv, bn
    conv, bn, relu
    conv, relu
    linear, bn
    linear, relu
    For these sequences, the first element in the output module list performs
    the fused operation. The rest of the elements are set to nn.Identity()
    """
    types = tuple(type_before_parametrizations(m) for m in mod_list)
    fuser_method = get_fuser_method(types, additional_fuser_method_mapping)
    if fuser_method is None:
        raise NotImplementedError("Cannot fuse modules: {}".format(types))
    new_mod : List[Optional[nn.Module]] = [None] * len(mod_list)
    fused = fuser_method(is_qat, *mod_list)
    # NOTE: forward hooks not processed in the two following for loops will be lost after the fusion
    # Move pre forward hooks of the base module to resulting fused module
    for handle_id, pre_hook_fn in mod_list[0]._forward_pre_hooks.items():
        fused.register_forward_pre_hook(pre_hook_fn)
        del mod_list[0]._forward_pre_hooks[handle_id]
    # Move post forward hooks of the last module to resulting fused module
    for handle_id, hook_fn in mod_list[-1]._forward_hooks.items():
        fused.register_forward_hook(hook_fn)
        del mod_list[-1]._forward_hooks[handle_id]
    new_mod[0] = fused
    for i in range(1, len(mod_list)):
        identity = nn.Identity()
        identity.training = mod_list[0].training
        new_mod[i] = identity
    return new_mod

DEFAULT_OP_LIST_TO_FUSER_METHOD: Dict[Tuple, Union[nn.Sequential, Callable]] = {
    (nn.Conv2d, nn.BatchNorm2d): fuse_conv_bn,
    (nn.Conv2d, nn.BatchNorm2d, nn.ReLU): fuse_conv_bn_relu,
    (nn.Conv2d, nn.ReLU): sequential_wrapper2(nni.ConvReLU2d),
    (nn.Linear, nn.BatchNorm1d): fuse_linear_bn,
    (nn.Linear, nn.ReLU): sequential_wrapper2(nni.LinearReLU),
    (nn.BatchNorm2d, nn.ReLU): sequential_wrapper2(nni.BNReLU2d),
}

def fuse_conv_bn(is_qat, conv, bn):
    assert(conv.training == bn.training),\
        "Conv and BN both must be in the same mode (train or eval)."
    fused_module_class_map = {
        nn.Conv1d: nni.ConvBn1d,
        nn.Conv2d: nni.ConvBn2d,
        nn.Conv3d: nni.ConvBn3d,
    }
    if is_qat:
        assert bn.num_features == conv.out_channels, 'Output channel of Conv2d must match num_features of BatchNorm2d'
        assert bn.affine, 'Only support fusing BatchNorm2d with affine set to True'
        assert bn.track_running_stats, 'Only support fusing BatchNorm2d with tracking_running_stats set to True'
        fused_module_class = fused_module_class_map.get((type(conv)), None)
        if fused_module_class is not None:
            return fused_module_class(conv, bn)
        else:
            raise NotImplementedError("Cannot fuse train modules: {}".format((conv, bn)))
    else:
        return nn.utils.fuse_conv_bn_eval(conv, bn)

def fuse_conv_bn_eval(conv, bn, transpose=False):
    assert(not (conv.training or bn.training)), "Fusion only for eval!"
    fused_conv = copy.deepcopy(conv)
    fused_conv.weight, fused_conv.bias = \
        fuse_conv_bn_weights(fused_conv.weight, fused_conv.bias,
                             bn.running_mean, bn.running_var, bn.eps, bn.weight, bn.bias, transpose)
    return fused_conv

print('\n Before fusion \n\n', r18_o.layer1)
r18_o.eval()
r18 = torch.quantization.fuse_modules(
    r18_o,
    [['conv1', 'bn1', 'relu'],
     ['layer1.0.conv1', 'layer1.0.bn1'], # , 'layer1.0.relu'],
     ['layer1.0.conv2', 'layer1.0.bn2'],
     ['layer1.1.conv1', 'layer1.1.bn1'], #, 'layer1.1.relu'],
     ['layer1.1.conv2', 'layer1.1.bn2']]
)
print('\n After fusion\n\n', r18.layer1)

def fuse_model(self):
    for m in self.modules():
        if type(m) == BasicBlock:
            torch.quantization.fuse_modules(m, [['conv1', 'bn1', 'relu'], ['conv2', 'bn2']], inplace=True)

def fuse_model(self):
    for m in self.modules():
        if type(m) == ConvBNReLU:
            torch.quantization.fuse_modacules(m, ['0', '1', '2'], inplace=True)
        if type(m) == InvertedResidual:
            for idx in range(len(m.conv)):
                if type(m.conv[idx]) == nn.Conv2d:
                    torch.quantization.fuse_modules(m.conv, [str(idx), str(idx + 1)], inplace=True)