-
Notifications
You must be signed in to change notification settings - Fork 85
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Cannot install on Windows #16
Comments
After some Googling, adding an environment variable Hopefully that helps other Windows users. |
So, I got through all the installs, ie.
When I run
I ran those 2 pip installs, ie
but I still get the same error? |
I tried to reproduce the errors mentioned above, but failed. Can you tell me the details of your running environment? For example, the operating system version, python seems to be 3.10? If this problem cannot be solved in the end, it is actually a better choice to directly use LM Studio to run the model, so that you can run larger models, and you do not need to wait for a long time each time you start AIlice. In addition, the voice dialogue function is not very good at the moment, if you are not interested in developing it, it is best to wait until this module is refactored before experiencing it. |
Windows 11 and Python 3.10.8.
No problem if it is not 100% ready yet. I can wait. One of my users requested I added support for it in Visions of Chaos. |
I forgot to remind you early that AIlice does not yet support Windows PowerShell scripts, so you will encounter problems when use the programming function of AIlice. I am trying to reproduce your problem. If it is solved successfully, I will consider adding PowerShell support. |
I installed AIlice in the conda environment under Windows, and tried programming and executing tasks. It turns out that WSL can execute these bash codes very well, so it seems that we do not really need to support powershell. But I still haven't reproduced these installation errors. Let me clean up AIlice's dependencies first. |
Maybe a requirements.txt with all dependancies like other repos would help? Then the install steps would be the git clone followed by pip install -r requirements.txt If you do go that was, please ensure version numbers for each package, ie No rush, take your time. |
OK, I will provide a requirements.txt after the voice dialogue function is perfected and useless dependencies are removed. Please wait for a few days. |
I just submitted a requirements.txt in the main branch, tested under python 3.10.13 environment. |
OK, I cloned the latest repo, created a new venv with
|
|
OK, got a bit further.
CUDA is definitely installed as I use it for many other Python AI/ML scripts. |
"Keyword arguments {'add_special_tokens': False} not recognized." "ASSISTANT_AIlice: D:\Tests\AIlice\voc_ailice\lib\site-packages\transformers\models\mistral\modeling_mistral.py:688: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at ..\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:263.) |
OK, my mistake. I was running some other Python code in the background so it was slower than expected. I do now get a response. I get error messages before the reponses, but it seems to work.
Getting closer. |
Yes, you can communicate with AIlice directly through voice conversation: Note that you may need to wait for a while when using it for the first time, AIlice needs to download model weights in the background. |
OK, thanks. Maybe add some text after "files started." when downloading the models... After the models download I can open the UI. I allow it access to my microphone.
When I play the recorded audio it is sped up 2 times and is repeated twice. If I type a text input the AI does respond and speak the response. |
Yeah, I'm thinking of adding a progress prompt as well, it's just that for some technical reasons this simple little task has become a bit difficult. If you are eager to try the voice conversation function, you can also use the voice conversation function of ailice_main, which provides a mechanism similar to natural dialogue between people, but is not as controllable as ailice_web. |
I can wait for a fix for the UI voice conversation fix. That is what people will want to use rather than the commandline. |
You may want to pay attention to this issue: |
OK, thanks for that link. It may be an issue with Firefox byu the looks of it. I was going to test, but... I tried a clean install today. Starting from a fresh environment, then git clone, then the pip installs. Now no matter how I do the installs (
That is at the point that the models would normally be downloaded? Is there a temp outage with the model server/hosts? |
AIlice's peripheral modules (including speech) are RPC services of independent processes. Some modules are necessary to start AIlice, if it fails to start, AIlice will fail. python3 -m ailice.modules.ASpeech --addr=ipc:///tmp/ASpeech.ipc In the future, I will add some error handling and user-friendly error reporting mechanisms to remind users. |
In addition, because the current timeout setting is too short, this error may occasionally occur. Just try again. |
I just submitted a change to the master branch to avoid "Resource temporarily unavailable" errors caused by slow startup of peripheral modules. You can update the code to see if the problem is solved. |
OK, testing again today. This works This still fails with "Resource temporarily unavailable" If I try
This is the batch file I now use to install a clean setup from scratch to test. Uses the original pip install -e methods and updates bitsandbytes to a working version and torch for GPU support.
|
Ah, this is a known issue. Some dependencies, maybe espnet, occasionally have versions that do not correctly reference its dependency typeguard, causing problems. |
Rolling back typeguard got it working.
|
Launching the gradio UI on Firefox has the bug that the recorded sound is twice as fast and doubled up. |
Thank you for your detailed information! |
If you just install "torch" on Windows you get the CPU version. You need to specify the GPU version by "torch==2.1.1+cu118". Then the other fixes so gradio and your scripts start. Most of my environments these days have some form of pip uninstall / pip installs at the end to fix version numbers or packages that have since been updated and broken older code. |
I create new clean Python environment.
Then the following commands all work
But when I try the next
pip install -e .[speech]
it fails with these errors...fail.txt
Any ideas? Thanks.
The text was updated successfully, but these errors were encountered: