No GPU detected/allowed, using most conservative search space.

Greetings. I am using windows 10, cuda version 10.1, python 3.8.5. GTX 1060 MaxQ. I installed autogluon as follow

pip install torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html
python -m pip install -U pip
python -m pip install -U setuptools wheel
python -m pip install -U "mxnet_cu101<2.0.0"
python -m pip install autogluon

I checked GPU presence as follow

import mxnet as mx
mx.context.num_gpus()

It output as 1

package info mxnet-cu101==1.5.0 autogluon==0.2.0 autogluon-contrib-nlp==0.0.1b20210201 autogluon.core==0.2.0 autogluon.extra==0.2.0 autogluon.features==0.2.0 autogluon.mxnet==0.2.0 autogluon.tabular==0.2.0 autogluon.text==0.2.0 autogluon.vision==0.2.0

But when I run predictor.fit it say INFO:gluoncv.auto.tasks.image_classification:No GPU detected/allowed, using most conservative search space.

Any idea about it?

[xxx2009-lab]

I also got the same issue @Innixma import mxnet as mx mx.context.num_gpus() 1

ag.get_gpu_count() 0

[Innixma]

@talhaanwarch latest pre-release should already be migrated to PyTorch via #1249

You should be able to run ImagePredictor without MXNet installed.

An example torch model to specify:

hyperparameters={'model': 'gluon_resnet101_v1d'}

You can see an example colab notebook that trains on Image without MXNet installed: https://colab.research.google.com/drive/1bQiruKftkwN43jTov09R8sjoXGdLdbuj?usp=sharing

Note: I’d generally recommend using TabularPredictor for Images, it has fewer bugs and better performance + benefits of ensembling.