GPU available but not used

[alevangel]

Executing CFLOW model on Colab, PyTorch returns this

GPU available: True, used: False
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs

GPU available but not used. Set "accelerator" and "devices" using Trainer(accelerator='gpu', devices=1).

Even if i'm setting gpus=1 on the config.yml of cflow.

To Reproduce
train.py --model cflow

GPU Model
GPU 0: Tesla T4

Versions

pytorch-lightning             1.6.1
torch                         1.10.0+cu111
torchaudio                    0.10.0+cu111
torchmetrics                  0.7.3
torchsummary                  1.5.1
torchtext                     0.11.0
torchvision                   0.11.1+cu111

[samet-akcay]

Hi @alevangel, Which GPU model do you have on Colab?

[alevangel]

Hi @alevangel, Which GPU model do you have on Colab?

GPU 0: Tesla T4 (edited)

[ashwinvaidya17]

@alevangel Can you share your torch version?

[alevangel]

@alevangel Can you share your torch version?

pytorch-lightning             1.6.1
torch                         1.10.0+cu111
torchaudio                    0.10.0+cu111
torchmetrics                  0.7.3
torchsummary                  1.5.1
torchtext                     0.11.0
torchvision                   0.11.1+cu111

[alevangel]

I solved rolling back at version of pytorch-lightning==1.5.9

[andriy-onufriyenko]

I have the same problem with CFLOW model on my local machine.

GPU available: True, used: False
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs

GPU available but not used. Set `accelerator` and `devices` using `Trainer(accelerator='gpu', devices=1)

python tools/train.py --model cflow

GPU:
NVIDIA GeForce GTX 1650 Ti

pytorch-lightning 1.6.1

[alevangel]

@andriy-onufriyenko use pytorch-lightning==1.5.9

[andriy-onufriyenko]

@andriy-onufriyenko use pytorch-lightning==1.5.9

@alevangel Install pytorch-lightning==1.5.9 and have another problem now:

ERROR: pip's dependency resolver does not currently take into account all the packages that are
installed. This behaviour is the source of the following dependency conflicts.
anomalib 0.2.6 requires pytorch-lightning>=1.6.0, but you have
pytorch-lightning 1.5.9 which is incompatible.

ModuleNotFoundError: No module named 'torchtext.legacy'

[samet-akcay]

@andriy-onufriyenko, can you add torch>=1.8.1 above this line, and do pip install -e again?

[andriy-onufriyenko]

@andriy-onufriyenko, can you add torch>=1.8.1 above this line, and do pip install -e again?

@samet-akcay

albumentations>=1.1.0
einops>=0.3.2
kornia>=0.5.6
omegaconf>=2.1.1
opencv-python>=4.5.3.56
pandas>=1.1.0
pytorch-lightning>=1.6.0
torch>=1.8.1
torchvision>=0.9.1
torchtext>=0.9.1
wandb==0.12.9
matplotlib>=3.4.3

It's working but without GPU

GPU available: True, used: False
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs

GPU available but not used. Set `accelerator` and `devices` using `Trainer(accelerator='gpu', devices=1)`

Trying edit config.yaml :

# PL Trainer Args. Don't add extra parameter here.
trainer:
  accelerator: 'gpu'

and got

ValueError: Unsupported accelerator found: gpu. Should be one of [null, ddp]

Parameter auto_select_gpus: true didn't affect.

[samet-akcay]

I solved rolling back at version of pytorch-lightning==1.5.9

I found the problem. PyTorch Lightning deprecated and / or modified some of the Trainer configs after v1.6.0, which is the reason for not using the GPU despite being available.

The issue has been fixed in PR #234

[andriy-onufriyenko]

I solved rolling back at version of pytorch-lightning==1.5.9

I found the problem. PyTorch Lightning deprecated and / or modified some of the Trainer configs after v1.6.0, which is the reason for not using the GPU despite being available.

The issue has been fixed in PR #234

@samet-akcay

I checked it works for me:

GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs

Just replace all 11 files from PR 234 in to my project.