{ "id": 209786, "title": "TF version difference between TPU and GPU", "url": "/competitions/cassava-leaf-disease-classification/discussion/209786", "author_name": "", "post_date": "2021-01-08T15:15:30.819509100Z", "votes": 4, "comment_count": 6, "views": 0, "content": "

I see the version of TF when I select GPU is 2.3.0 and with TPU it is 2.2.0. Any reason why? I want to use efficientNet on TPU, but it is not available in that version.

", "messages": [ { "id": "1144625", "postDate": "01/08/2021 15:15:30", "content": "

I see the version of TF when I select GPU is 2.3.0 and with TPU it is 2.2.0. Any reason why? I want to use efficientNet on TPU, but it is not available in that version.

", "rawMarkdown": "I see the version of TF when I select GPU is 2.3.0 and with TPU it is 2.2.0. Any reason why? I want to use efficientNet on TPU, but it is not available in that version.", "votes": null }, { "id": "1144711", "postDate": "01/08/2021 16:17:23", "content": "

Maybe the 2.3.0 version doesn't support TPU execution.

", "rawMarkdown": "Maybe the **2.3.0** version doesn't support TPU execution.", "votes": null }, { "id": "1146632", "postDate": "01/09/2021 22:29:51", "content": "

EfficientNet is only available on tf-nightly.
\nTry this

", "rawMarkdown": "EfficientNet is only available on tf-nightly.\nTry [this](https://pypi.org/project/efficientnet/)", "votes": null }, { "id": "1147895", "postDate": "01/10/2021 19:41:14", "content": "

you can update the Tensorflow version easily in TPU. You can see this kernel where I update Tensorflow versions in TPU and also use efficient net. Though I used external installations Efficient net, You can use Tensorflow Keras application efficient net.

", "rawMarkdown": "you can update the Tensorflow version easily in TPU. You can [see this kernel ](https://www.kaggle.com/durbin164/tpu-bitempered-logistic-loss-keras-tensorflow) where I update Tensorflow versions in TPU and also use efficient net. Though I used external installations Efficient net, You can use Tensorflow Keras application efficient net.", "votes": null }, { "id": "1148766", "postDate": "01/11/2021 11:24:14", "content": "

Thanks Masud 👍

", "rawMarkdown": "Thanks Masud 👍", "votes": null }, { "id": "1149042", "postDate": "01/11/2021 14:47:33", "content": "

In the past, the TF version 2.3 had memory issues so at the time the Kaggle team kept version 2.2 for TPUs, I am not sure if version 2.3 is working already.

", "rawMarkdown": "In the past, the TF version 2.3 had memory issues so at the time the Kaggle team kept version 2.2 for TPUs, I am not sure if version 2.3 is working already.", "votes": null }, { "id": "1149583", "postDate": "01/12/2021 01:20:05", "content": "

tensorflow performs differently on TPU and GPU, some operations cannot be used on TPU, please seehttps://cloud.google.com/tpu/docs/tensorflow-ops

", "rawMarkdown": "tensorflow performs differently on TPU and GPU, some operations cannot be used on TPU, please see[https://cloud.google.com/tpu/docs/tensorflow-ops](url)", "votes": null } ], "comments": [ { "id": 1144711, "author_name": "hiramcho", "author_url": "", "post_date": "01/08/2021 16:17:23", "content": "

Maybe the 2.3.0 version doesn't support TPU execution.

", "votes": null, "replies": [] }, { "id": 1146632, "author_name": "mohitve", "author_url": "", "post_date": "01/09/2021 22:29:51", "content": "

EfficientNet is only available on tf-nightly.
\nTry this

", "votes": null, "replies": [] }, { "id": 1147895, "author_name": "durbin164", "author_url": "", "post_date": "01/10/2021 19:41:14", "content": "

you can update the Tensorflow version easily in TPU. You can see this kernel where I update Tensorflow versions in TPU and also use efficient net. Though I used external installations Efficient net, You can use Tensorflow Keras application efficient net.

", "votes": null, "replies": [ { "id": 1148766, "author_name": "spiraldesign", "author_url": "", "post_date": "01/11/2021 11:24:14", "content": "

Thanks Masud 👍

", "votes": null, "replies": [] } ] }, { "id": 1149042, "author_name": "dimitreoliveira", "author_url": "", "post_date": "01/11/2021 14:47:33", "content": "

In the past, the TF version 2.3 had memory issues so at the time the Kaggle team kept version 2.2 for TPUs, I am not sure if version 2.3 is working already.

", "votes": null, "replies": [] }, { "id": 1149583, "author_name": "zhangeng", "author_url": "", "post_date": "01/12/2021 01:20:05", "content": "

tensorflow performs differently on TPU and GPU, some operations cannot be used on TPU, please seehttps://cloud.google.com/tpu/docs/tensorflow-ops

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1144625": "I see the version of TF when I select GPU is 2.3.0 and with TPU it is 2.2.0. Any reason why? I want to use efficientNet on TPU, but it is not available in that version.", "1144711": "Maybe the **2.3.0** version doesn't support TPU execution.", "1146632": "EfficientNet is only available on tf-nightly.\nTry [this](https://pypi.org/project/efficientnet/)", "1147895": "you can update the Tensorflow version easily in TPU. You can [see this kernel ](https://www.kaggle.com/durbin164/tpu-bitempered-logistic-loss-keras-tensorflow) where I update Tensorflow versions in TPU and also use efficient net. Though I used external installations Efficient net, You can use Tensorflow Keras application efficient net.", "1148766": "Thanks Masud 👍", "1149042": "In the past, the TF version 2.3 had memory issues so at the time the Kaggle team kept version 2.2 for TPUs, I am not sure if version 2.3 is working already.", "1149583": "tensorflow performs differently on TPU and GPU, some operations cannot be used on TPU, please see[https://cloud.google.com/tpu/docs/tensorflow-ops](url)" }, "source": "meta" }