TypeError: 'staticmethod' object is not callable

[pin24]

Windows 10
Python 3.8
torch-directml-0.2.4.dev240815

2.3.1+cpu
Traceback (most recent call last):
File "tt_v6.py", line 17, in
import torch_directml
File "C:\Users\Zver.conda\envs\gan\lib\site-packages\torch_directml_init_.py", line 27, in
class PrivateUse1Module:
File "C:\Users\Zver.conda\envs\gan\lib\site-packages\torch_directml_init_.py", line 74, in PrivateUse1Module
def has_float64_support(device_id = default_device()):
TypeError: 'staticmethod' object is not callable

[brazeiro63]

I've experienced the same problem with version 2.4.1+cpu.
the methods has_float64_support and gpu_memory definitions are calling default_device() directly.
This is causing an error "'staticmethod' object is not callable".

My workaroud:

    @staticmethod
    def has_float64_support(device_id = None):
        if device_id is None:
            device_id = PrivateUse1Module.default_device()
        return torch_directml_native.has_float64_support(device_id)

    @staticmethod
    def gpu_memory(device_id = None, mb_per_tile = 1):
        if device_id is None:
            device_id = PrivateUse1Module.default_device()
        return torch_directml_native.get_gpu_memory(device_id, mb_per_tile)

[devrix123]

I also got this. what I did was to change the init.py code in the torch_directml library and comment out the @staticmethod code for the default_device() function. after that, I successfully imported the torch_directml library

#@staticmethod
    def default_device() -> int:
        r"""Returns the index of the default selected device."""
        return torch_directml_native.get_default_device()

I honestly don't know if what I'm doing is wrong or not.

[k3sh4vk]

@pin24 @brazeiro63 @devrix123

The example seems to be working fine in the following configuration without any workarounds -
Python Version = 3.10.11
torch_directml Version = 0.2.5.dev240914

But fails if we try to use python >= 3.7 and < 3.10

It seems that the code in new versions of the torch-directml library is heavily dependent on the new advancements related to staticmethod and classmethod. POINT NO.9 introduced from Python 3.10 onwards.

Here's the PR which introduced this feature.

[danila-permogorskii]

@pin24 @brazeiro63 @devrix123

The example seems to be working fine in the following configuration without any workarounds - Python Version = 3.10.11 torch_directml Version = 0.2.5.dev240914

But fails if we try to use python >= 3.7 and < 3.10

It seems that the code in new versions of the torch-directml library is heavily dependent on the new advancements related to staticmethod and classmethod. POINT NO.9 introduced from Python 3.10 onwards.

Here's the PR which introduced this feature.

Yeah, it works! Tack så mycket!