RuntimeError: Library cublas64_12.dll is not found or cannot be loaded

[DT05YT]

i tried this, and first attempt at it resulted in this:

(.venv) C:\Users\me\Downloads\UltraSinger-main\src>UltraSinger.py -i "song.mp3"

[UltraSinger] *****************************
[UltraSinger] UltraSinger Version: 0.0.9
[UltraSinger] *****************************
[UltraSinger] Checking GPU support for tensorflow and pytorch.
[UltraSinger] tensorflow - using cuda gpu.
[UltraSinger] pytorch - using cuda gpu.
[UltraSinger] full automatic mode
[UltraSinger] Searching song in musicbrainz
[UltraSinger] cant find title whip in song
[UltraSinger] No match found
[UltraSinger] Creating output folder. -> output\song (1)
[UltraSinger] Creating output folder. -> output\song (1)\cache
[UltraSinger] Separating vocals from audio with demucs and cuda as worker.
Important: the default model was recently changed to `htdemucs` the latest Hybrid Transformer Demucs model. In some cases, this model can actually perform worse than previous models. To get back the old default model use `-n mdx_extra_q`.
Selected model is a bag of 1 models. You will see that many progress bars per track.
Separated tracks will be stored in C:\Users\me\Downloads\UltraSinger-main\src\separated\htdemucs
Separating track output\song (1)\song.mp3
100%|██████████████████████████████████████████████████████████████████████| 181.35/181.35 [00:06<00:00, 26.18seconds/s]
[UltraSinger] Converting wav to mp3
[UltraSinger] Reduce noise from vocal audio with ffmpeg.
[UltraSinger] Converting audio for AI
[UltraSinger] Mute audio parts with no singing
[UltraSinger] Loading whisper with model large-v2 and cuda as worker
No language specified, language will be first be detected for each audio file (increases inference time).
Lightning automatically upgraded your loaded checkpoint from v1.5.4 to v2.2.1. To apply the upgrade to your files permanently, run `python -m pytorch_lightning.utilities.upgrade_checkpoint C:\Users\me\.cache\torch\whisperx-vad-segmentation.bin`
Model was trained with pyannote.audio 0.0.1, yours is 3.1.1. Bad things might happen unless you revert pyannote.audio to 0.x.
Model was trained with torch 1.10.0+cu102, yours is 2.0.1+cu117. Bad things might happen unless you revert torch to 1.x.
[UltraSinger] Transcribing output\song (1)\cache\song_mute.wav
Traceback (most recent call last):
  File "C:\Users\me\Downloads\UltraSinger-main\src\UltraSinger.py", line 984, in <module>
    main(sys.argv[1:])
  File "C:\Users\me\Downloads\UltraSinger-main\src\UltraSinger.py", line 883, in main
    run()
  File "C:\Users\me\Downloads\UltraSinger-main\src\UltraSinger.py", line 392, in run
    detected_language, transcribed_data = transcribe_audio()
  File "C:\Users\me\Downloads\UltraSinger-main\src\UltraSinger.py", line 515, in transcribe_audio
    transcribed_data, detected_language = transcribe_with_whisper(
  File "C:\Users\me\Downloads\UltraSinger-main\src\modules\Speech_Recognition\Whisper.py", line 61, in transcribe_with_whisper
    result = loaded_whisper_model.transcribe(
  File "C:\Users\me\Downloads\UltraSinger-main\.venv\lib\site-packages\whisperx\asr.py", line 194, in transcribe
    language = language or self.detect_language(audio)
  File "C:\Users\me\Downloads\UltraSinger-main\.venv\lib\site-packages\whisperx\asr.py", line 252, in detect_language
    encoder_output = self.model.encode(segment)
  File "C:\Users\me\Downloads\UltraSinger-main\.venv\lib\site-packages\whisperx\asr.py", line 86, in encode
    return self.model.encode(features, to_cpu=to_cpu)
RuntimeError: Library cublas64_12.dll is not found or cannot be loaded

What does it all mean

[Etidou]

Hey, @rakuri255 and @DT05YT I had exactly the same errors have you found a solution ?

I'm with Windows 11 OS with a new installation (I tried on 2 computers and got the same error, I tried with a youtube link and a direct .mp3 file).

[jonathanoh-94]

Hello,

I wanted to report that I'm facing a similar issue as described in this thread, particularly with Nvidia CUDA v11.8. It appears that the core of the problem might be related to compatibility or configuration issues with CUDA.

I am currently in the process of testing a workaround by forcing the operations to run on the CPU instead, using the --force_cpu True option, in hopes of bypassing the GPU acceleration issues tied to CUDA v11.8. I have not yet concluded if this approach will fully resolve the issue, but I wanted to share my ongoing experience with the community.

If anyone has encountered a similar situation and has insights, advice, or potential fixes, it would be greatly appreciated. I am eager to find a solution that allows for the smooth operation of UltraSinger without the hurdles posed by CUDA v11.8.

[ponelll]

Hello,
I had the same issue with CUDA v11.8. After updating CUDA to v12.4, it works perfectly. I guess that's because the missing dll is cublas64_12.dll. CUDA v11.8 only has cublas64_11.dll.

[rakuri255]

Hi,

@Etidou , @DT05YT , @jonathanoh-94 , @ponelll can you post your GPUs?
Did you install PyTorch with cu117, cu118 or cu121?

[Etidou]

Hey @rakuri255 My GPU is a Nvidia GEForce GTX 1080 and i'm running with cu117 !

[jonathanoh-94]

Hi,

@Etidou , @DT05YT , @jonathanoh-94 , @ponelll can you post your GPUs? Did you install PyTorch with cu117, cu118 or cu121?

After updating to v12.4, it worked! Thank you !

[rakuri255]

I think the WhisperX trunk had some updates that breaks with pytorch 11.x

Is it solved for you all with cu121?

[Etidou]

It works for me too!

I just updated pip install torch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 --index-url https://download.pytorch.org/whl/cu121 as you mentioned everything works fine now.

Thanks !

[rakuri255]

It works for me too!

I just updated pip install torch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 --index-url https://download.pytorch.org/whl/cu121 as you mentioned everything works fine now.

Thanks !

Isnt this CPU only?

[rakuri255]

When i use the latest cu121 with my RTX2080 i get Error: cublas64_11.dll instead of Error: cublas64_12.dll