Closed geekinchief22 closed 1 year ago
same problem here
@geekinchief22 @HMS97 I had a similar problem (cuDNN version error) and was able to fix it by doing the following:
requirements.txt
to unpin versions -- should look like this:
wget
nemo_toolkit[asr]
transformers>=4.26.1
faster-whisper
. . .
pip install -r requirements.txt
). For me, this replaced my PyTorch 2.1 with an older version.Manually install a newer nightly PyTorch on top of the repo:
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu118
Thought I'd share this in case it helps you folk. I'm using the NVIDIA NGC PyTorch containers (23.04-py3) with driver version 525.85.05
(Ubuntu 22.04 as the host OS--the containers run Ubuntu 20.04). PyTorch is supposed to come bundled with CUDA/cuDNN, but I'm guessing that for me the error was coming from a version mismatch between the newer torch version I already had, and the older one the repo's requirements wanted to install.
I managed to solve the problem by removing the cudnn package that was installed in pip and torch automatically used the original cudnn installation that was installed using the official instructions
I get this every time I try to run it. I've tried like 100x ways to get the new cuDNN. I am using Miniconda, but nothing works. Any ideas?
RuntimeError: cuDNN version incompatibility: PyTorch was compiled against (8, 5, 0) but found runtime version (8, 2, 1). PyTorch already comes bundled with cuDNN. One option to resolving this error is to ensure PyTorch can find the bundled cuDNN.