Can't "prefer_gpu()"

[Mathhead]

Hi there.

I am trying to run spaCy on my GPU, yet "prefer_gpu()" returns False.

How to reproduce the behaviour

!pip install -U 'spacy[cuda114]'
import spacy
spacy.prefer_gpu()

Your Environment

2022-03-03 15:48:07.197671: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libnvinfer.so.6
2022-03-03 15:48:07.201202: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libnvinfer_plugin.so.6

Info about spaCy

• spaCy version: 3.2.3
• Platform: Linux-5.4.0-99-generic-x86_64-with-Ubuntu-18.04-bionic
• Python version: 3.6.9
  -**CUDA 11.4
  -**Driver Version: 470.63.01
  -**tf.config.list_physical_devices(): [PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU'),
  PhysicalDevice(name='/physical_device:XLA_CPU:0', device_type='XLA_CPU'),
  PhysicalDevice(name='/physical_device:XLA_GPU:0', device_type='XLA_GPU'),
  PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

[Mathhead]

Hi persons.

Could you please give me any advice on what information I could deliver to help you help me?

I am trying to run spacy on a virtual GPU. Does this help you?

[polm]

Check these imports:

import cupy
import cupyx
import cupy.cuda
from cupy.cuda.compiler import compile_with_cache
import torch

If any of those fail, that could cause this.

Also I note you are using Python 3.6. I think that should still work in theory, but it did reach end-of-life last year; I recommend upgrading to more recent version.

[Mathhead]

Thank you.

It seems that I have problems with cupy.

I cannot uninstall it properly, so I think, there is something horribly wrong with my installation. Therefore, I figure, I should sort out my problem with cupy first.