Check out the release notes for major features and improvements such as a new Profiler for TF 2 or updates on tf.distribute and tf.keras.
\n\nhttps://github.com/tensorflow/tensorflow/releases/tag/v2.2.0-rc4
\n\n", "messages": [ { "id": "830959", "postDate": "05/03/2020 02:31:29", "content": "Check out the release notes for major features and improvements such as a new Profiler for TF 2 or updates on tf.distribute and tf.keras.
\n\nhttps://github.com/tensorflow/tensorflow/releases/tag/v2.2.0-rc4
\n\n", "rawMarkdown": "Check out the release notes for major features and improvements such as a new Profiler for TF 2 or updates on tf.distribute and tf.keras. \n\nhttps://github.com/tensorflow/tensorflow/releases/tag/v2.2.0-rc4\n\n[Source](https://github.com/tensorflow/tensorflow/releases/tag/v2.2.0-rc4)", "votes": null }, { "id": "832189", "postDate": "05/04/2020 00:09:44", "content": "Do you know if kaggle will switch from tf 2.1 to tf 2.2 soon ? I noticed on colab that tf 2.2 fixes the keras fit class_weight bug
", "rawMarkdown": "Do you know if kaggle will switch from tf 2.1 to tf 2.2 soon ? I noticed on colab that tf 2.2 fixes the keras fit class_weight bug", "votes": null }, { "id": "832263", "postDate": "05/04/2020 02:21:01", "content": "@mgornergoogle maybe the best person to comment on this
", "rawMarkdown": "mgornergoogle maybe the best person to comment on this", "votes": null }, { "id": "832328", "postDate": "05/04/2020 04:09:13", "content": "I'll ask him. Thank you.
", "rawMarkdown": "I'll ask him. Thank you.", "votes": null }, { "id": "834750", "postDate": "05/05/2020 18:44:39", "content": "Not before there is a TF 2.2 release. It's only an RC4 for now. But yes, I agree with you, there are a couple of useful bug fixes in TF 2.2 for TPUs, especially around stability. And using tf-nightly with TPUs is not an option unfortunately. All I can do is ask for a little bit more patience.
", "rawMarkdown": "Not before there is a TF 2.2 release. It's only an RC4 for now. But yes, I agree with you, there are a couple of useful bug fixes in TF 2.2 for TPUs, especially around stability. And using tf-nightly with TPUs is not an option unfortunately. All I can do is ask for a little bit more patience.", "votes": null }, { "id": "834753", "postDate": "05/05/2020 18:46:42", "content": "ok thanks for your reply.
", "rawMarkdown": "ok thanks for your reply.", "votes": null }, { "id": "835524", "postDate": "05/06/2020 10:21:46", "content": "@mgornergoogle \" And using tf-nightly with TPUs is not an option unfortunately...\"
\n\nIs there a rule? Because one can install it and for sure it makes a difference using your custom training loop.
\n\n@albangabillon the class_weight bug can easily be worked around because you can call the function with a weight array as far as I can remember.
", "rawMarkdown": "mgornergoogle \" And using tf-nightly with TPUs is not an option unfortunately...\"\n\nIs there a rule? Because one can install it and for sure it makes a difference using your custom training loop.\n\n@albangabillon the class_weight bug can easily be worked around because you can call the function with a weight array as far as I can remember.", "votes": null }, { "id": "836069", "postDate": "05/06/2020 17:37:06", "content": "pip installing tf-nightly only installs it on your Kaggle VM. The TPU part does not get updated. This might lead to miscommunication between the two and weird bugs or error. You might get lucky but it is not recommended.
", "rawMarkdown": "pip installing tf-nightly only installs it on your Kaggle VM. The TPU part does not get updated. This might lead to miscommunication between the two and weird bugs or error. You might get lucky but it is not recommended.", "votes": null }, { "id": "836113", "postDate": "05/06/2020 18:18:08", "content": "@romanweilguny \nyes you can use an array. It does not raise an error, but from what I observed, the array is simply ignored.
", "rawMarkdown": "romanweilguny \nyes you can use an array. It does not raise an error, but from what I observed, the array is simply ignored.", "votes": null }, { "id": "836303", "postDate": "05/06/2020 21:55:44", "content": "here we go :)
", "rawMarkdown": "here we go :)", "votes": null }, { "id": "837792", "postDate": "05/08/2020 03:45:52", "content": "So TF 2.2 now released and i've confirmed it on my device. I just wonder if Kaggle will switch to TF 2.2 before competition is over.
\n\n```\nPython 3.8.2 (default, Apr 1 2020, 15:52:55) \n[GCC 9.3.0] on linux\nType \"help\", \"copyright\", \"credits\" or \"license\" for more information.
\n\n\n", "rawMarkdown": "So TF 2.2 now released and i've confirmed it on my device. I just wonder if Kaggle will switch to TF 2.2 before competition is over.\n\n```\nPython 3.8.2 (default, Apr 1 2020, 15:52:55) \n[GCC 9.3.0] on linux\nType \"help\", \"copyright\", \"credits\" or \"license\" for more information.\n>>> import tensorflow as tf\n>>> tf.__version__\n'2.2.0'\n```", "votes": null }, { "id": "839028", "postDate": "05/09/2020 02:35:10", "content": "\n\n\n\nimport tensorflow as tf\n tf.version\n '2.2.0'\n ```
\n
I was wondering as well. Can we use 2.2 for our latest submissions?
", "rawMarkdown": "I was wondering as well. Can we use 2.2 for our latest submissions?", "votes": null }, { "id": "845586", "postDate": "05/13/2020 09:50:15", "content": "@mgornergoogle \nAs the TF 2.2 is no longer in RC state when can we expect upgrade of Kaggle environment?
", "rawMarkdown": "mgornergoogle \nAs the TF 2.2 is no longer in RC state when can we expect upgrade of Kaggle environment?", "votes": null }, { "id": "846364", "postDate": "05/13/2020 17:53:33", "content": "We are working on updating Kaggle's image to TF2.2 - I'd expect ~1-2 weeks for the latest release to become the default for TF.
", "rawMarkdown": "We are working on updating Kaggle's image to TF2.2 - I'd expect ~1-2 weeks for the latest release to become the default for TF.", "votes": null }, { "id": "846453", "postDate": "05/13/2020 19:10:45", "content": "Thanks for the update @juliaelliott
", "rawMarkdown": "Thanks for the update @juliaelliott", "votes": null }, { "id": "863603", "postDate": "05/27/2020 12:29:57", "content": "Any updates on the TF2.2?
", "rawMarkdown": "Any updates on the TF2.2?", "votes": null }, { "id": "876352", "postDate": "06/06/2020 16:31:08", "content": "Any update on TF2.2?
", "rawMarkdown": "Any update on TF2.2?", "votes": null }, { "id": "878833", "postDate": "06/08/2020 22:34:47", "content": "Hi everyone - sorry for the delay. The rollout has begun, so it should hit 100% of sessions over the next few days. You can pip show tensorflow to check which version your session is running.
Do you know if kaggle will switch from tf 2.1 to tf 2.2 soon ? I noticed on colab that tf 2.2 fixes the keras fit class_weight bug
", "votes": null, "replies": [ { "id": 832263, "author_name": "kurianbenoy", "author_url": "", "post_date": "05/04/2020 02:21:01", "content": "@mgornergoogle maybe the best person to comment on this
", "votes": null, "replies": [] }, { "id": 832328, "author_name": "albangabillon", "author_url": "", "post_date": "05/04/2020 04:09:13", "content": "I'll ask him. Thank you.
", "votes": null, "replies": [] }, { "id": 834750, "author_name": "mgorner", "author_url": "", "post_date": "05/05/2020 18:44:39", "content": "Not before there is a TF 2.2 release. It's only an RC4 for now. But yes, I agree with you, there are a couple of useful bug fixes in TF 2.2 for TPUs, especially around stability. And using tf-nightly with TPUs is not an option unfortunately. All I can do is ask for a little bit more patience.
", "votes": null, "replies": [] }, { "id": 834753, "author_name": "albangabillon", "author_url": "", "post_date": "05/05/2020 18:46:42", "content": "ok thanks for your reply.
", "votes": null, "replies": [] }, { "id": 845586, "author_name": "rzepinskip", "author_url": "", "post_date": "05/13/2020 09:50:15", "content": "@mgornergoogle \nAs the TF 2.2 is no longer in RC state when can we expect upgrade of Kaggle environment?
", "votes": null, "replies": [] }, { "id": 846364, "author_name": "juliaelliott", "author_url": "", "post_date": "05/13/2020 17:53:33", "content": "We are working on updating Kaggle's image to TF2.2 - I'd expect ~1-2 weeks for the latest release to become the default for TF.
", "votes": null, "replies": [] }, { "id": 846453, "author_name": "kurianbenoy", "author_url": "", "post_date": "05/13/2020 19:10:45", "content": "Thanks for the update @juliaelliott
", "votes": null, "replies": [] }, { "id": 863603, "author_name": "tolgadincer", "author_url": "", "post_date": "05/27/2020 12:29:57", "content": "Any updates on the TF2.2?
", "votes": null, "replies": [] }, { "id": 876352, "author_name": "ahmedabdelali", "author_url": "", "post_date": "06/06/2020 16:31:08", "content": "Any update on TF2.2?
", "votes": null, "replies": [] }, { "id": 878833, "author_name": "juliaelliott", "author_url": "", "post_date": "06/08/2020 22:34:47", "content": "Hi everyone - sorry for the delay. The rollout has begun, so it should hit 100% of sessions over the next few days. You can pip show tensorflow to check which version your session is running.
@mgornergoogle \" And using tf-nightly with TPUs is not an option unfortunately...\"
\n\nIs there a rule? Because one can install it and for sure it makes a difference using your custom training loop.
\n\n@albangabillon the class_weight bug can easily be worked around because you can call the function with a weight array as far as I can remember.
", "votes": null, "replies": [ { "id": 836069, "author_name": "mgorner", "author_url": "", "post_date": "05/06/2020 17:37:06", "content": "pip installing tf-nightly only installs it on your Kaggle VM. The TPU part does not get updated. This might lead to miscommunication between the two and weird bugs or error. You might get lucky but it is not recommended.
", "votes": null, "replies": [] }, { "id": 836113, "author_name": "albangabillon", "author_url": "", "post_date": "05/06/2020 18:18:08", "content": "@romanweilguny \nyes you can use an array. It does not raise an error, but from what I observed, the array is simply ignored.
", "votes": null, "replies": [] }, { "id": 836303, "author_name": "romanweilguny", "author_url": "", "post_date": "05/06/2020 21:55:44", "content": "here we go :)
", "votes": null, "replies": [] } ] }, { "id": 837792, "author_name": "ilosvigil", "author_url": "", "post_date": "05/08/2020 03:45:52", "content": "So TF 2.2 now released and i've confirmed it on my device. I just wonder if Kaggle will switch to TF 2.2 before competition is over.
\n\n```\nPython 3.8.2 (default, Apr 1 2020, 15:52:55) \n[GCC 9.3.0] on linux\nType \"help\", \"copyright\", \"credits\" or \"license\" for more information.
\n\n\n", "votes": null, "replies": [ { "id": 839028, "author_name": "fabmilo", "author_url": "", "post_date": "05/09/2020 02:35:10", "content": "\n\n\n\nimport tensorflow as tf\n tf.version\n '2.2.0'\n ```
\n
I was wondering as well. Can we use 2.2 for our latest submissions?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "830959": "Check out the release notes for major features and improvements such as a new Profiler for TF 2 or updates on tf.distribute and tf.keras. \n\nhttps://github.com/tensorflow/tensorflow/releases/tag/v2.2.0-rc4\n\n[Source](https://github.com/tensorflow/tensorflow/releases/tag/v2.2.0-rc4)", "832189": "Do you know if kaggle will switch from tf 2.1 to tf 2.2 soon ? I noticed on colab that tf 2.2 fixes the keras fit class_weight bug", "832263": "mgornergoogle maybe the best person to comment on this", "832328": "I'll ask him. Thank you.", "834750": "Not before there is a TF 2.2 release. It's only an RC4 for now. But yes, I agree with you, there are a couple of useful bug fixes in TF 2.2 for TPUs, especially around stability. And using tf-nightly with TPUs is not an option unfortunately. All I can do is ask for a little bit more patience.", "834753": "ok thanks for your reply.", "835524": "mgornergoogle \" And using tf-nightly with TPUs is not an option unfortunately...\"\n\nIs there a rule? Because one can install it and for sure it makes a difference using your custom training loop.\n\n@albangabillon the class_weight bug can easily be worked around because you can call the function with a weight array as far as I can remember.", "836069": "pip installing tf-nightly only installs it on your Kaggle VM. The TPU part does not get updated. This might lead to miscommunication between the two and weird bugs or error. You might get lucky but it is not recommended.", "836113": "romanweilguny \nyes you can use an array. It does not raise an error, but from what I observed, the array is simply ignored.", "836303": "here we go :)", "837792": "So TF 2.2 now released and i've confirmed it on my device. I just wonder if Kaggle will switch to TF 2.2 before competition is over.\n\n```\nPython 3.8.2 (default, Apr 1 2020, 15:52:55) \n[GCC 9.3.0] on linux\nType \"help\", \"copyright\", \"credits\" or \"license\" for more information.\n>>> import tensorflow as tf\n>>> tf.__version__\n'2.2.0'\n```", "839028": "I was wondering as well. Can we use 2.2 for our latest submissions?", "845586": "mgornergoogle \nAs the TF 2.2 is no longer in RC state when can we expect upgrade of Kaggle environment?", "846364": "We are working on updating Kaggle's image to TF2.2 - I'd expect ~1-2 weeks for the latest release to become the default for TF.", "846453": "Thanks for the update @juliaelliott", "863603": "Any updates on the TF2.2?", "876352": "Any update on TF2.2?", "878833": "Hi everyone - sorry for the delay. The rollout has begun, so it should hit 100% of sessions over the next few days. You can `pip show tensorflow` to check which version your session is running." }, "source": "meta" }