Windows 10 installation problem

[PasF]

Describe your environment

• GPU: [cuda]
• VRAM: [12GB]
• CPU arch: [x86]
• OS: [Windows 10]
• Python: [Anaconda]
• Branch: [main]
• Commit: [commit 03f5ebb (HEAD -> main, origin/main, origin/HEAD)
  Author: spezialspezial 75758219+spezialspezial@users.noreply.github.com
  Date: Wed Oct 26 15:18:07 2022 +0200]

Describe the bug
Tried to install thanks to conda and pyenv on Windows 10, same error every time:

(invokeai) C:\Users\w00t\Desktop\ai_art\InvokeAI>python scripts\preload_models.py
NOTE: Redirects are currently not supported in Windows or MacOs.
Traceback (most recent call last):
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\transformers\utils\import_utils.py", line 1002, in _get_module
    return importlib.import_module("." + module_name, self.__name__)
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\importlib\__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1050, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1027, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1006, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 688, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 883, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\transformers\models\clip\modeling_clip.py", line 27, in <module>
    from ...modeling_utils import PreTrainedModel
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\transformers\modeling_utils.py", line 79, in <module>
    from accelerate import dispatch_model, infer_auto_device_map, init_empty_weights
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\accelerate\__init__.py", line 7, in <module>
    from .accelerator import Accelerator
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\accelerate\accelerator.py", line 27, in <module>
    from .checkpointing import load_accelerator_state, load_custom_state, save_accelerator_state, save_custom_state
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\accelerate\checkpointing.py", line 24, in <module>
    from .utils import (
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\accelerate\utils\__init__.py", line 96, in <module>
    from .launch import PrepareForLaunch, _filter_args, get_launch_prefix
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\accelerate\utils\launch.py", line 25, in <module>
    import torch.distributed.run as distrib_run
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\run.py", line 386, in <module>
    from torch.distributed.launcher.api import LaunchConfig, elastic_launch
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\launcher\__init__.py", line 10, in <module>
    from torch.distributed.launcher.api import (  # noqa: F401
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\launcher\api.py", line 15, in <module>
    from torch.distributed.elastic.agent.server.api import WorkerSpec
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\elastic\agent\server\__init__.py", line 40, in <module>
    from .local_elastic_agent import TORCHELASTIC_ENABLE_FILE_TIMER, TORCHELASTIC_TIMER_FILE
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\elastic\agent\server\local_elastic_agent.py", line 19, in <module>
    import torch.distributed.elastic.timer as timer
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\elastic\timer\__init__.py", line 44, in <module>
    from .file_based_local_timer import FileTimerClient, FileTimerServer, FileTimerRequest  # noqa: F401
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\elastic\timer\file_based_local_timer.py", line 63, in <module>
    class FileTimerClient(TimerClient):
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\torch\distributed\elastic\timer\file_based_local_timer.py", line 81, in FileTimerClient
    def __init__(self, file_path: str, signal=signal.SIGKILL) -> None:
AttributeError: module 'signal' has no attribute 'SIGKILL'. Did you mean: 'SIGILL'?

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "C:\Users\w00t\Desktop\ai_art\InvokeAI\scripts\preload_models.py", line 6, in <module>
    from transformers import CLIPTokenizer, CLIPTextModel
  File "<frozen importlib._bootstrap>", line 1075, in _handle_fromlist
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\transformers\utils\import_utils.py", line 993, in __getattr__
    value = getattr(module, name)
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\transformers\utils\import_utils.py", line 992, in __getattr__
    module = self._get_module(self._class_to_module[name])
  File "C:\Users\w00t\anaconda3\envs\invokeai\lib\site-packages\transformers\utils\import_utils.py", line 1004, in _get_module
    raise RuntimeError(
RuntimeError: Failed to import transformers.models.clip.modeling_clip because of the following error (look up to see its traceback):
module 'signal' has no attribute 'SIGKILL'

To Reproduce
Steps to reproduce the behavior:

1. Follow https://github.com/invoke-ai/InvokeAI/blob/main/docs/installation/INSTALL_WINDOWS.md
2. Run python scripts\preload_models.py on step 7

Expected behavior
I don't know but not an error.

[n00mkrad]

Replacing SIGKILL with SIGTERM might work.

[PasF]

It worked, changed \invokeai\lib\site-packages\torch\distributed\elastic\timer\file_based_local_timer.py

def __init__(self, file_path: str, signal=signal.SIGKILL)
to
def __init__(self, file_path: str, signal=signal.SIGTERM)

Thanks !

[UnicodeTreason]

Confirmed, hacky workaround works.
What's the root cause here?
I assume sometime in the last few weeks a newer version of pytorch has been added which includes this SIGKILL bug.
I'm outta my depth here a little but maybe:
65b1468
Which has allowed
pytorch/pytorch#85427
To squeeze its way in.

[UnicodeTreason]

These updated modules also caused my install to fail to detect CUDA entirely.
Forcing my environment.yml back to the following (And rebuilding the cuda environment) resolved both issues for me:

dependencies:
  - python>=3.9
  - pip>=22.2
  - cudatoolkit=11.3
  - pytorch=1.11.0
  - torchvision=0.12.0
  - numpy=1.19.2

[GifCo]

I can confirm I have the same issue on Win 10 with Cuda. Fresh install with pew or with Conda both fail with the module 'signal' has no attribute 'SIGKILL'

I am able to get a working install thanks to UnicodeTreason by adding:

torch==1.11.0

to requirments-lin-win-colab-CUDA.txt

Edit: also need to specify CudaToolkit otherwise torch will use CPU. Was unable to get CU 11.6 so had to downgrade to CU 11.3. Make sure you also change the --extra-index-url to point to v 11.3

--extra-index-url https://download.pytorch.org/whl/cu113

torch==1.12.0+cu113

[n00mkrad]

Seems like a Pytorch developer accidentally added a Linux-only command into a script.
SIGTERM (terminate) is basically what a program receives when you send CTRL+C to cancel something in a CLI.
Linux has some more like SIGTERM, but those do not work on Windows.

I guess for now users will have to rollback and/or wait for a fix. Or, of course, patch it themselves (you can just find-and-replace signal.SIGKILL with signal.SIGTERM and it should work).

[sveken]

Same issue on Windows 11,
Changing to SIGTERM resolved it

[bkonkle]

Changing the lines in requirements-lin-win-colab-CUDA.txt to this worked for me on Windows 11 in PowerShell with Python 3.10 installed via the Microsoft Store and using pew for my virtualenv:

--extra-index-url https://download.pytorch.org/whl/cu116 --trusted-host https://download.pytorch.org
torch==1.12.1+cu116

Then I activated my virtual environment:

pew workon invokeai

And ran pip to update dependencies:

pip install -r requirements-lin-win-colab-CUDA.txt

[desis123]

It worked, changed \invokeai\lib\site-packages\torch\distributed\elastic\timer\file_based_local_timer.py

def __init__(self, file_path: str, signal=signal.SIGKILL) to def __init__(self, file_path: str, signal=signal.SIGTERM)

Thanks !

I was having the same issue from another project, and your solution worked like a charm, it saved my days .

[psychedelicious]

We sorted this issue last week. I don't recall the specific PR or version, but at least in the upcoming v2.2 this is sorted.