In the Getting Started notebook, there are some FLAGS about passing a TPU address to the notebook.
Currently I don't quite understand what is the interplay between the Kaggle Submission Kernels and the GCP services like TPU or GCP Storage Buckets. Could you help me, sort this out:
\n\nAs for the GCP Storage, I guess I can link my GCP account, and this would authorize my notebook for accessing a GCP storage bucket, right? But if I pass a TPU address how and what would authorize the notebook to use the TPU?
\n\nOr is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?
", "messages": [ { "id": "676731", "postDate": "11/19/2019 13:19:52", "content": "In the Getting Started notebook, there are some FLAGS about passing a TPU address to the notebook.
Currently I don't quite understand what is the interplay between the Kaggle Submission Kernels and the GCP services like TPU or GCP Storage Buckets. Could you help me, sort this out:
\n\nAs for the GCP Storage, I guess I can link my GCP account, and this would authorize my notebook for accessing a GCP storage bucket, right? But if I pass a TPU address how and what would authorize the notebook to use the TPU?
\n\nOr is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?
", "rawMarkdown": "In the Getting Started notebook, there are some `FLAGS` about passing a TPU address to the notebook.\n\nCurrently I don't quite understand what is the interplay between the Kaggle Submission Kernels and the GCP services like TPU or GCP Storage Buckets. Could you help me, sort this out:\n\n1. When I commit a kernel it will be executed and generate a submission file. During this \"commit\" execution of the kernel, is there any way of using a TPU or accessing a GCP storage bucket (containing a trained checkpoint for example)?\n2. Or are there any TPU resources that would be used by the evaluation script during the commit?\n\nAs for the GCP Storage, I guess I can link my GCP account, and this would authorize my notebook for accessing a GCP storage bucket, right? But if I pass a TPU address how and what would authorize the notebook to use the TPU?\n\nOr is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?", "votes": null }, { "id": "676801", "postDate": "11/19/2019 14:10:44", "content": "\n\n\nOr is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?
\n
I think that's the case, I have not heard about TPU support in kaggle kernels, although I've heard it's coming.
", "rawMarkdown": "> Or is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?\n\nI think that's the case, I have not heard about TPU support in kaggle kernels, although I've heard it's coming.", "votes": null }, { "id": "683368", "postDate": "11/28/2019 10:55:56", "content": "Anyway, for submitting to competition, only CPU or GPU code is accepted: https://www.kaggle.com/c/tensorflow2-question-answering/overview/code-requirements
", "rawMarkdown": "Anyway, for submitting to competition, only CPU or GPU code is accepted: https://www.kaggle.com/c/tensorflow2-question-answering/overview/code-requirements", "votes": null } ], "comments": [ { "id": 676801, "author_name": "lopuhin", "author_url": "", "post_date": "11/19/2019 14:10:44", "content": "\n\n\nOr is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?
\n
I think that's the case, I have not heard about TPU support in kaggle kernels, although I've heard it's coming.
", "votes": null, "replies": [] }, { "id": 683368, "author_name": "arvindpdmn", "author_url": "", "post_date": "11/28/2019 10:55:56", "content": "Anyway, for submitting to competition, only CPU or GPU code is accepted: https://www.kaggle.com/c/tensorflow2-question-answering/overview/code-requirements
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "676731": "In the Getting Started notebook, there are some `FLAGS` about passing a TPU address to the notebook.\n\nCurrently I don't quite understand what is the interplay between the Kaggle Submission Kernels and the GCP services like TPU or GCP Storage Buckets. Could you help me, sort this out:\n\n1. When I commit a kernel it will be executed and generate a submission file. During this \"commit\" execution of the kernel, is there any way of using a TPU or accessing a GCP storage bucket (containing a trained checkpoint for example)?\n2. Or are there any TPU resources that would be used by the evaluation script during the commit?\n\nAs for the GCP Storage, I guess I can link my GCP account, and this would authorize my notebook for accessing a GCP storage bucket, right? But if I pass a TPU address how and what would authorize the notebook to use the TPU?\n\nOr is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?", "676801": "> Or is the TPU related code in the notebook just copy paste from the original example, that is never meant to execute during evaluation as the Kaggle Notebooks do not support TPU?\n\nI think that's the case, I have not heard about TPU support in kaggle kernels, although I've heard it's coming.", "683368": "Anyway, for submitting to competition, only CPU or GPU code is accepted: https://www.kaggle.com/c/tensorflow2-question-answering/overview/code-requirements" }, "source": "meta" }