![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4576004%2F26efd23492fd131b316c218bd8da2dd3%2Ferror.PNG?generation=1594386451007133&alt=media\")
After defining the neural net architecture with EfficientNet B0 as the backbone and compiling it, the GPU memory suddenly becomes full which makes it impossible to train the model.\nI'm attaching a screenshot of the same.
Can anyone help ?
", "messages": [ { "id": 922985, "postDate": "2020-07-10T13:14:15.300Z", "content": "You can try to restart the kernel and do it again. Or you use smaller images.
", "rawMarkdown": "You can try to restart the kernel and do it again. Or you use smaller images.", "replies": [ { "id": 923002, "postDate": "2020-07-10T13:27:06.433Z", "content": "@derinformatiker Hey, thanks for the fast reply. \nThat's the problem though before executing this very cell the gpu memory was hardly occupied 350 mb but just after model architecture and compiling (without even starting training), the memory gets full.\n I've tried restarting the kernel several time, all in vain. Do you have any idea?
", "rawMarkdown": "@derinformatiker Hey, thanks for the fast reply. \nThat's the problem though before executing this very cell the gpu memory was hardly occupied 350 mb but just after model architecture and compiling (without even starting training), the memory gets full.\n I've tried restarting the kernel several time, all in vain. Do you have any idea? " }, { "id": 923053, "postDate": "2020-07-10T13:57:27.087Z", "content": "I don´t know why i can´t reproduce this error. Maybe its the strategy.scope ?
", "rawMarkdown": " I don´t know why i can´t reproduce this error. Maybe its the strategy.scope ?" }, { "id": 923238, "postDate": "2020-07-10T16:30:47.593Z", "content": "@derinformatiker but it doesn't make any sense to do with scope, right?
", "rawMarkdown": "@derinformatiker but it doesn't make any sense to do with scope, right? " } ] }, { "id": 922975, "postDate": "2020-07-10T13:07:50.670Z", "content": "After defining the neural net architecture with EfficientNet B0 as the backbone and compiling it, the GPU memory suddenly becomes full which makes it impossible to train the model.\nI'm attaching a screenshot of the same.
Can anyone help ?
", "rawMarkdown": "After defining the neural net architecture with EfficientNet B0 as the backbone and compiling it, the GPU memory suddenly becomes full which makes it impossible to train the model.\nI'm attaching a screenshot of the same.\n\nCan anyone help ?" } ], "comments": [ { "id": 922985, "author_name": "Der Informatiker", "author_url": "", "post_date": "2020-07-10T13:14:15.300000", "content": "You can try to restart the kernel and do it again. Or you use smaller images.
", "votes": 0, "replies": [ { "id": 923002, "author_name": "Nikhil Bartwal", "author_url": "", "post_date": "2020-07-10T13:27:06.433000", "content": "@derinformatiker Hey, thanks for the fast reply. \nThat's the problem though before executing this very cell the gpu memory was hardly occupied 350 mb but just after model architecture and compiling (without even starting training), the memory gets full.\n I've tried restarting the kernel several time, all in vain. Do you have any idea?
", "votes": 0, "replies": [] }, { "id": 923053, "author_name": "Der Informatiker", "author_url": "", "post_date": "2020-07-10T13:57:27.087000", "content": "I don´t know why i can´t reproduce this error. Maybe its the strategy.scope ?
", "votes": 0, "replies": [] }, { "id": 923238, "author_name": "Nikhil Bartwal", "author_url": "", "post_date": "2020-07-10T16:30:47.593000", "content": "@derinformatiker but it doesn't make any sense to do with scope, right?
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "922985": "You can try to restart the kernel and do it again. Or you use smaller images.", "922975": "After defining the neural net architecture with EfficientNet B0 as the backbone and compiling it, the GPU memory suddenly becomes full which makes it impossible to train the model.\nI'm attaching a screenshot of the same.\n\nCan anyone help ?" } }