torch.inference_mode() not working with torch_directml device

[nagadomi]

torch_directml==0.2.2.dev240614

import torch
import torch_directml

device = torch_directml.device()
model = torch.nn.Linear(3, 128).eval().to(device)
with torch.inference_mode():
    x = torch.zeros(4, 3).to(device)
    z = model(x)

RuntimeError: Cannot set version_counter for inference tensor

Using torch.no_grad() instead of torch.inference_mode() works fine.