{ "id": 188845, "title": "If you cannot find GPU after using kaggle_l5kit and using torch", "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/188845", "author_name": "", "post_date": "2020-10-05T15:35:12.061676Z", "votes": 1, "comment_count": 1, "views": 0, "content": "

In Pytorch's official website the correct way to install torch via pip with cuda support is:

\n
pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html\n
\n

since both Google Colab and Kaggle base docker images only has 10.1. If you only do

\n
pip install torch==1.6.0\n
\n

it's only compatible with cuda 10.2. I guess that's the real reason. I'll try to get https://github.com/Kaggle/docker-python/pull/883/files merged and fix that if possible. I guess in the meantime we'll have to be careful using or installing torch version which isn't built-in.

\n

To see the cuda version:

\n

\"\"

", "messages": [ { "id": "1038151", "postDate": "10/05/2020 15:35:12", "content": "

In Pytorch's official website the correct way to install torch via pip with cuda support is:

\n
pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html\n
\n

since both Google Colab and Kaggle base docker images only has 10.1. If you only do

\n
pip install torch==1.6.0\n
\n

it's only compatible with cuda 10.2. I guess that's the real reason. I'll try to get https://github.com/Kaggle/docker-python/pull/883/files merged and fix that if possible. I guess in the meantime we'll have to be careful using or installing torch version which isn't built-in.

\n

To see the cuda version:

\n

\"\"

", "rawMarkdown": "In Pytorch's [official website](https://pytorch.org/get-started/locally/) the correct way to install torch via `pip` with cuda support is:\n\n```\npip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html\n```\n\nsince both Google Colab and Kaggle base docker images only has 10.1. If you only do\n\n```\npip install torch==1.6.0\n```\n\nit's only compatible with cuda 10.2. I guess that's the real reason. I'll try to get https://github.com/Kaggle/docker-python/pull/883/files merged and fix that if possible. I guess in the meantime we'll have to be careful using or installing torch version which isn't built-in.\n\n_To see the cuda version:_\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1138256%2Fc18627059410419f83f1d38b9a15f433%2FScreen%20Shot%202020-10-05%20at%2023.35.53.png?generation=1601912195967732&alt=media)", "votes": null }, { "id": "1038159", "postDate": "10/05/2020 15:42:55", "content": "

I've had one successful example here: https://www.kaggle.com/etareduce/environment-setup-with-torch-1-6-l5kit-1-1-0-gpu

", "rawMarkdown": "I've had one successful example here: https://www.kaggle.com/etareduce/environment-setup-with-torch-1-6-l5kit-1-1-0-gpu", "votes": null } ], "comments": [ { "id": 1038159, "author_name": "etareduce", "author_url": "", "post_date": "10/05/2020 15:42:55", "content": "

I've had one successful example here: https://www.kaggle.com/etareduce/environment-setup-with-torch-1-6-l5kit-1-1-0-gpu

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1038151": "In Pytorch's [official website](https://pytorch.org/get-started/locally/) the correct way to install torch via `pip` with cuda support is:\n\n```\npip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html\n```\n\nsince both Google Colab and Kaggle base docker images only has 10.1. If you only do\n\n```\npip install torch==1.6.0\n```\n\nit's only compatible with cuda 10.2. I guess that's the real reason. I'll try to get https://github.com/Kaggle/docker-python/pull/883/files merged and fix that if possible. I guess in the meantime we'll have to be careful using or installing torch version which isn't built-in.\n\n_To see the cuda version:_\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1138256%2Fc18627059410419f83f1d38b9a15f433%2FScreen%20Shot%202020-10-05%20at%2023.35.53.png?generation=1601912195967732&alt=media)", "1038159": "I've had one successful example here: https://www.kaggle.com/etareduce/environment-setup-with-torch-1-6-l5kit-1-1-0-gpu" }, "source": "meta" }