{ "id": 411547, "title": "Time out error on very simple model", "url": "/competitions/birdclef-2023/discussion/411547", "author_name": "", "post_date": "2023-05-19T16:52:51.831010800Z", "votes": 1, "comment_count": 6, "views": 0, "content": "

Can any one tell me how I get notebook time out when submitting a notebook using this simple model??
\nam I making something wrong??

\n

class TinyModel(torch.nn.Module):
\n def init(self, numChannels, classes):
\n # call the parent constructor
\n super(TinyModel, self).init()
\n # initialize first set of CONV => RELU => POOL layers
\n self.conv1 = Conv2d(in_channels=3, out_channels=20,
\n kernel_size=(5, 5))
\n self.relu1 = ReLU()
\n self.maxpool1 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
\n # initialize second set of CONV => RELU => POOL layers
\n self.conv2 = Conv2d(in_channels=20, out_channels=50,
\n kernel_size=(5, 5))
\n self.relu2 = ReLU()
\n self.maxpool2 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))

\n
    self.conv3 = Conv2d(in_channels=50, out_channels=20,\n        kernel_size=(1, 1))\n    self.relu3 = ReLU()\n\n    self.fc1 = Linear(in_features=56180, out_features=5000)\n    self.relu3 = ReLU()\n\n    self.fc2 = Linear(in_features=5000, out_features=500)\n    self.relu2 = ReLU()\n\n    # initialize our softmax classifier\n    self.fc3 = Linear(in_features=500, out_features=classes)\n    self.logSoftmax = LogSoftmax(dim=1)\n\ndef forward(self, x):\n    # pass the input through our first set of CONV => RELU =>\n    # POOL layers\n    x = self.conv1(x)\n    x = self.relu1(x)\n    x = self.maxpool1(x)\n    # pass the output from the previous layer through the second\n    # set of CONV => RELU => POOL layers\n    x = self.conv2(x)\n    x = self.relu2(x)\n    x = self.maxpool2(x)\n\n    x = self.conv3(x)\n    x = self.relu3(x)\n    #x = self.\n    # flatten the output from the previous layer and pass it\n    # through our only set of FC => RELU layers\n    x = flatten(x, 1)\n    x = self.fc1(x)\n    x = self.relu3(x)\n\n    x = self.fc2(x)\n    x = self.relu3(x)\n    # pass the output to our softmax classifier to get our output\n    # predictions\n    x = self.fc3(x)\n    output = self.logSoftmax(x)\n    # return the output predictions\n    return output\n
", "messages": [ { "id": "2266032", "postDate": "05/19/2023 16:52:51", "content": "

Can any one tell me how I get notebook time out when submitting a notebook using this simple model??
\nam I making something wrong??

\n

class TinyModel(torch.nn.Module):
\n def init(self, numChannels, classes):
\n # call the parent constructor
\n super(TinyModel, self).init()
\n # initialize first set of CONV => RELU => POOL layers
\n self.conv1 = Conv2d(in_channels=3, out_channels=20,
\n kernel_size=(5, 5))
\n self.relu1 = ReLU()
\n self.maxpool1 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
\n # initialize second set of CONV => RELU => POOL layers
\n self.conv2 = Conv2d(in_channels=20, out_channels=50,
\n kernel_size=(5, 5))
\n self.relu2 = ReLU()
\n self.maxpool2 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))

\n
    self.conv3 = Conv2d(in_channels=50, out_channels=20,\n        kernel_size=(1, 1))\n    self.relu3 = ReLU()\n\n    self.fc1 = Linear(in_features=56180, out_features=5000)\n    self.relu3 = ReLU()\n\n    self.fc2 = Linear(in_features=5000, out_features=500)\n    self.relu2 = ReLU()\n\n    # initialize our softmax classifier\n    self.fc3 = Linear(in_features=500, out_features=classes)\n    self.logSoftmax = LogSoftmax(dim=1)\n\ndef forward(self, x):\n    # pass the input through our first set of CONV => RELU =>\n    # POOL layers\n    x = self.conv1(x)\n    x = self.relu1(x)\n    x = self.maxpool1(x)\n    # pass the output from the previous layer through the second\n    # set of CONV => RELU => POOL layers\n    x = self.conv2(x)\n    x = self.relu2(x)\n    x = self.maxpool2(x)\n\n    x = self.conv3(x)\n    x = self.relu3(x)\n    #x = self.\n    # flatten the output from the previous layer and pass it\n    # through our only set of FC => RELU layers\n    x = flatten(x, 1)\n    x = self.fc1(x)\n    x = self.relu3(x)\n\n    x = self.fc2(x)\n    x = self.relu3(x)\n    # pass the output to our softmax classifier to get our output\n    # predictions\n    x = self.fc3(x)\n    output = self.logSoftmax(x)\n    # return the output predictions\n    return output\n
", "rawMarkdown": "Can any one tell me how I get notebook time out when submitting a notebook using this simple model??\nam I making something wrong??\n\nclass TinyModel(torch.nn.Module):\n def __init__(self, numChannels, classes):\n # call the parent constructor\n super(TinyModel, self).__init__()\n # initialize first set of CONV => RELU => POOL layers\n self.conv1 = Conv2d(in_channels=3, out_channels=20,\n kernel_size=(5, 5))\n self.relu1 = ReLU()\n self.maxpool1 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))\n # initialize second set of CONV => RELU => POOL layers\n self.conv2 = Conv2d(in_channels=20, out_channels=50,\n kernel_size=(5, 5))\n self.relu2 = ReLU()\n self.maxpool2 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))\n \n self.conv3 = Conv2d(in_channels=50, out_channels=20,\n kernel_size=(1, 1))\n self.relu3 = ReLU()\n \n self.fc1 = Linear(in_features=56180, out_features=5000)\n self.relu3 = ReLU()\n \n self.fc2 = Linear(in_features=5000, out_features=500)\n self.relu2 = ReLU()\n \n # initialize our softmax classifier\n self.fc3 = Linear(in_features=500, out_features=classes)\n self.logSoftmax = LogSoftmax(dim=1)\n\n def forward(self, x):\n # pass the input through our first set of CONV => RELU =>\n # POOL layers\n x = self.conv1(x)\n x = self.relu1(x)\n x = self.maxpool1(x)\n # pass the output from the previous layer through the second\n # set of CONV => RELU => POOL layers\n x = self.conv2(x)\n x = self.relu2(x)\n x = self.maxpool2(x)\n \n x = self.conv3(x)\n x = self.relu3(x)\n #x = self.\n # flatten the output from the previous layer and pass it\n # through our only set of FC => RELU layers\n x = flatten(x, 1)\n x = self.fc1(x)\n x = self.relu3(x)\n \n x = self.fc2(x)\n x = self.relu3(x)\n # pass the output to our softmax classifier to get our output\n # predictions\n x = self.fc3(x)\n output = self.logSoftmax(x)\n # return the output predictions\n return output", "votes": null }, { "id": "2266117", "postDate": "05/19/2023 18:09:38", "content": "

The error most likely doesn't come from your model but maybe the way you use it since there is no reason such a model ends up in a timeout, we would need to see more of your code to properly help you

", "rawMarkdown": "The error most likely doesn't come from your model but maybe the way you use it since there is no reason such a model ends up in a timeout, we would need to see more of your code to properly help you", "votes": null }, { "id": "2266277", "postDate": "05/19/2023 21:36:54", "content": "

thank you for your reply
\nhere is the whole notebook https://colab.research.google.com/drive/1zR6AONcWBgClHr0yud0KxeEfSt87WUrf?usp=sharing, yes you pointed it right (since there is no reason such a model ends up in a timeout,) I saw people make advanced models and they work

", "rawMarkdown": "thank you for your reply\nhere is the whole notebook https://colab.research.google.com/drive/1zR6AONcWBgClHr0yud0KxeEfSt87WUrf?usp=sharing, yes you pointed it right (since there is no reason such a model ends up in a timeout,) I saw people make advanced models and they work", "votes": null }, { "id": "2267256", "postDate": "05/20/2023 18:14:02", "content": "

Hey @mohamedantargad,
\nI recommand that instead of taking each 5 second segment like you are doing (so a tensor of shape (1,num_channels,audio_lenght,frequences)) and then predict for every single one of those, you should concatenate them first into a (120,num_channels,audio_lenght,frequences) and then predict as if it was a batch of size 120.
\nThat might be the reason why your code doesn't run, you'll need to make a few tweaks but overall I don't spot any mistake other than that. Good luck.

", "rawMarkdown": "Hey @mohamedantargad,\nI recommand that instead of taking each 5 second segment like you are doing (so a tensor of shape `(1,num_channels,audio_lenght,frequences)`) and then predict for every single one of those, you should concatenate them first into a `(120,num_channels,audio_lenght,frequences)` and then predict as if it was a batch of size 120.\nThat might be the reason why your code doesn't run, you'll need to make a few tweaks but overall I don't spot any mistake other than that. Good luck.", "votes": null }, { "id": "2267668", "postDate": "05/21/2023 06:48:06", "content": "

Thank you again, I edited the notebook to make the batch of size 120 but still the same timeout error is thrown:
\nhttps://colab.research.google.com/drive/1w2soNvRE7JPNUqp7uOPowTCYmz1TJnSj?usp=sharing

", "rawMarkdown": "Thank you again, I edited the notebook to make the batch of size 120 but still the same timeout error is thrown:\nhttps://colab.research.google.com/drive/1w2soNvRE7JPNUqp7uOPowTCYmz1TJnSj?usp=sharing", "votes": null }, { "id": "2267841", "postDate": "05/21/2023 09:18:12", "content": "

Assuming the training process went smoothly, I recommend you to check out this thread:
\nAvoid Timeout by Rounding Your Model's Parameters

\n

There is a known issue about extremely long runtime when inferencing with denormals in the model parameters. Any solution in the thread, although slightly different, should all work.

", "rawMarkdown": "Assuming the training process went smoothly, I recommend you to check out this thread:\n[Avoid Timeout by Rounding Your Model's Parameters](https://www.kaggle.com/competitions/birdclef-2023/discussion/409331)\n\nThere is a known issue about extremely long runtime when inferencing with denormals in the model parameters. Any solution in the thread, although slightly different, should all work.", "votes": null }, { "id": "2269273", "postDate": "05/22/2023 10:49:51", "content": "

thanks a lot, I tried one of the solutions and it worked perfectly, now I'm trying a larger model
\nIt's my first competition and I've already loved the community
\nthanks again for you help

", "rawMarkdown": "thanks a lot, I tried one of the solutions and it worked perfectly, now I'm trying a larger model\nIt's my first competition and I've already loved the community\nthanks again for you help", "votes": null } ], "comments": [ { "id": 2266117, "author_name": "janmpia", "author_url": "", "post_date": "05/19/2023 18:09:38", "content": "

The error most likely doesn't come from your model but maybe the way you use it since there is no reason such a model ends up in a timeout, we would need to see more of your code to properly help you

", "votes": null, "replies": [ { "id": 2266277, "author_name": "mohamedantargad", "author_url": "", "post_date": "05/19/2023 21:36:54", "content": "

thank you for your reply
\nhere is the whole notebook https://colab.research.google.com/drive/1zR6AONcWBgClHr0yud0KxeEfSt87WUrf?usp=sharing, yes you pointed it right (since there is no reason such a model ends up in a timeout,) I saw people make advanced models and they work

", "votes": null, "replies": [ { "id": 2267256, "author_name": "janmpia", "author_url": "", "post_date": "05/20/2023 18:14:02", "content": "

Hey @mohamedantargad,
\nI recommand that instead of taking each 5 second segment like you are doing (so a tensor of shape (1,num_channels,audio_lenght,frequences)) and then predict for every single one of those, you should concatenate them first into a (120,num_channels,audio_lenght,frequences) and then predict as if it was a batch of size 120.
\nThat might be the reason why your code doesn't run, you'll need to make a few tweaks but overall I don't spot any mistake other than that. Good luck.

", "votes": null, "replies": [ { "id": 2267668, "author_name": "mohamedantargad", "author_url": "", "post_date": "05/21/2023 06:48:06", "content": "

Thank you again, I edited the notebook to make the batch of size 120 but still the same timeout error is thrown:
\nhttps://colab.research.google.com/drive/1w2soNvRE7JPNUqp7uOPowTCYmz1TJnSj?usp=sharing

", "votes": null, "replies": [] } ] } ] } ] }, { "id": 2267841, "author_name": "lhanhsin", "author_url": "", "post_date": "05/21/2023 09:18:12", "content": "

Assuming the training process went smoothly, I recommend you to check out this thread:
\nAvoid Timeout by Rounding Your Model's Parameters

\n

There is a known issue about extremely long runtime when inferencing with denormals in the model parameters. Any solution in the thread, although slightly different, should all work.

", "votes": null, "replies": [ { "id": 2269273, "author_name": "mohamedantargad", "author_url": "", "post_date": "05/22/2023 10:49:51", "content": "

thanks a lot, I tried one of the solutions and it worked perfectly, now I'm trying a larger model
\nIt's my first competition and I've already loved the community
\nthanks again for you help

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2266032": "Can any one tell me how I get notebook time out when submitting a notebook using this simple model??\nam I making something wrong??\n\nclass TinyModel(torch.nn.Module):\n def __init__(self, numChannels, classes):\n # call the parent constructor\n super(TinyModel, self).__init__()\n # initialize first set of CONV => RELU => POOL layers\n self.conv1 = Conv2d(in_channels=3, out_channels=20,\n kernel_size=(5, 5))\n self.relu1 = ReLU()\n self.maxpool1 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))\n # initialize second set of CONV => RELU => POOL layers\n self.conv2 = Conv2d(in_channels=20, out_channels=50,\n kernel_size=(5, 5))\n self.relu2 = ReLU()\n self.maxpool2 = MaxPool2d(kernel_size=(2, 2), stride=(2, 2))\n \n self.conv3 = Conv2d(in_channels=50, out_channels=20,\n kernel_size=(1, 1))\n self.relu3 = ReLU()\n \n self.fc1 = Linear(in_features=56180, out_features=5000)\n self.relu3 = ReLU()\n \n self.fc2 = Linear(in_features=5000, out_features=500)\n self.relu2 = ReLU()\n \n # initialize our softmax classifier\n self.fc3 = Linear(in_features=500, out_features=classes)\n self.logSoftmax = LogSoftmax(dim=1)\n\n def forward(self, x):\n # pass the input through our first set of CONV => RELU =>\n # POOL layers\n x = self.conv1(x)\n x = self.relu1(x)\n x = self.maxpool1(x)\n # pass the output from the previous layer through the second\n # set of CONV => RELU => POOL layers\n x = self.conv2(x)\n x = self.relu2(x)\n x = self.maxpool2(x)\n \n x = self.conv3(x)\n x = self.relu3(x)\n #x = self.\n # flatten the output from the previous layer and pass it\n # through our only set of FC => RELU layers\n x = flatten(x, 1)\n x = self.fc1(x)\n x = self.relu3(x)\n \n x = self.fc2(x)\n x = self.relu3(x)\n # pass the output to our softmax classifier to get our output\n # predictions\n x = self.fc3(x)\n output = self.logSoftmax(x)\n # return the output predictions\n return output", "2266117": "The error most likely doesn't come from your model but maybe the way you use it since there is no reason such a model ends up in a timeout, we would need to see more of your code to properly help you", "2266277": "thank you for your reply\nhere is the whole notebook https://colab.research.google.com/drive/1zR6AONcWBgClHr0yud0KxeEfSt87WUrf?usp=sharing, yes you pointed it right (since there is no reason such a model ends up in a timeout,) I saw people make advanced models and they work", "2267256": "Hey @mohamedantargad,\nI recommand that instead of taking each 5 second segment like you are doing (so a tensor of shape `(1,num_channels,audio_lenght,frequences)`) and then predict for every single one of those, you should concatenate them first into a `(120,num_channels,audio_lenght,frequences)` and then predict as if it was a batch of size 120.\nThat might be the reason why your code doesn't run, you'll need to make a few tweaks but overall I don't spot any mistake other than that. Good luck.", "2267668": "Thank you again, I edited the notebook to make the batch of size 120 but still the same timeout error is thrown:\nhttps://colab.research.google.com/drive/1w2soNvRE7JPNUqp7uOPowTCYmz1TJnSj?usp=sharing", "2267841": "Assuming the training process went smoothly, I recommend you to check out this thread:\n[Avoid Timeout by Rounding Your Model's Parameters](https://www.kaggle.com/competitions/birdclef-2023/discussion/409331)\n\nThere is a known issue about extremely long runtime when inferencing with denormals in the model parameters. Any solution in the thread, although slightly different, should all work.", "2269273": "thanks a lot, I tried one of the solutions and it worked perfectly, now I'm trying a larger model\nIt's my first competition and I've already loved the community\nthanks again for you help" }, "source": "meta" }