![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1062256%2F489ee4c87dcd6d837c1fe7655b3ba127%2FScreenshot%20from%202025-03-19%2022-43-55.png?generation=1742438724207925&alt=media\")
I am a beginner in Kaggle competitions and did not see any test datasets. Therefore, I copied a sample submission but was unable to upload it. The system informed me that 120 has exceeded the limit of 90
", "messages": [ { "id": "3154459", "postDate": "03/20/2025 01:08:46", "content": "I am a beginner in Kaggle competitions and did not see any test datasets. Therefore, I copied a sample submission but was unable to upload it. The system informed me that 120 has exceeded the limit of 90
", "rawMarkdown": "I am a beginner in Kaggle competitions and did not see any test datasets. Therefore, I copied a sample submission but was unable to upload it. The system informed me that 120 has exceeded the limit of 90", "votes": null }, { "id": "3154484", "postDate": "03/20/2025 02:28:58", "content": "Need a bit more info - the 120 suggests the notebook ran that many minutes, while the limit for the competition is 90. So you got to cut at least 30 minutes out.
\nNot sure I understand the rest of your issues. Share what you copied?
", "rawMarkdown": "Need a bit more info - the 120 suggests the notebook ran that many minutes, while the limit for the competition is 90. So you got to cut at least 30 minutes out.\n\nNot sure I understand the rest of your issues. Share what you copied?", "votes": null }, { "id": "3154488", "postDate": "03/20/2025 02:32:17", "content": "Your Notebook's runtime of 9 minutes exceeds this competition's GPU max of 1 minute. This is the issue I encountered. I don't know how to solve it
", "rawMarkdown": "Your Notebook's runtime of 9 minutes exceeds this competition's GPU max of 1 minute. This is the issue I encountered. I don't know how to solve it", "votes": null }, { "id": "3154498", "postDate": "03/20/2025 02:42:42", "content": "For prediction on test the notebook not suppose to use GPU.
\n\n\nGPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.>
\n
So disable GPU as a start point. The Accelerator drop down will let you select None - right now this probably shows one of the GPU's setups that can be used.
\nSince I do most my work on home pc I like to use three notebooks. The first one builds the data set. This competition has lots of options for doing things with the data. The 2nd notebook trains, while the third is for prediction. The third is the only one that I will upload to kaggle and run. My models are saved locally, so I will create a kaggle data set to add them to the prediction notebook on kaggle.
", "rawMarkdown": "For prediction on test the notebook not suppose to use GPU.\n\n>GPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.>\n\nSo disable GPU as a start point. The Accelerator drop down will let you select None - right now this probably shows one of the GPU's setups that can be used.\n\nSince I do most my work on home pc I like to use three notebooks. The first one builds the data set. This competition has lots of options for doing things with the data. The 2nd notebook trains, while the third is for prediction. The third is the only one that I will upload to kaggle and run. My models are saved locally, so I will create a kaggle data set to add them to the prediction notebook on kaggle.", "votes": null }, { "id": "3154501", "postDate": "03/20/2025 02:45:29", "content": "
Does this mean that training is a file, while testing should be pulled out as a separate notebook and used on the CPU?
", "rawMarkdown": "Does this mean that training is a file, while testing should be pulled out as a separate notebook and used on the CPU?", "votes": null }, { "id": "3154513", "postDate": "03/20/2025 02:59:09", "content": "Follow the example of several shared notebooks. They will generally train in one notebook and save the models. That can be done using GPU for speed.
\nThan a second notebook used to predict and submit. It must not use GPU. You add the saved models as data - a little tricky for new folks ……
", "rawMarkdown": "Follow the example of several shared notebooks. They will generally train in one notebook and save the models. That can be done using GPU for speed. \n\nThan a second notebook used to predict and submit. It must not use GPU. You add the saved models as data - a little tricky for new folks ......", "votes": null }, { "id": "3154520", "postDate": "03/20/2025 03:05:10", "content": "oh I roughly understand what you mean. I need to first train the model and save it to my local computer, then open Jupyter and upload my trained model as a prediction. This process cannot use GPU, is right?
", "rawMarkdown": "oh I roughly understand what you mean. I need to first train the model and save it to my local computer, then open Jupyter and upload my trained model as a prediction. This process cannot use GPU, is right?", "votes": null }, { "id": "3154529", "postDate": "03/20/2025 03:13:52", "content": "Think your almost saying it right. Train a model anywhere you want. Local machine, goggle colab, borrow Elon's 200K super compter, etc. Save the MODELS - look at the pytorch shared notebooks. They save 5 folds worth of models and than use those 5 models to make the predictions.
\nYou Kaggle notebook only needs the models - in the case of a pytorch that would be *.pth files. I create a kaggle data set to hold those and add it to my inference notebook as data. Again, the pytorch shared example shows the result - you just need to learn how to do the data add :)
", "rawMarkdown": "Think your almost saying it right. Train a model anywhere you want. Local machine, goggle colab, borrow Elon's 200K super compter, etc. Save the MODELS - look at the pytorch shared notebooks. They save 5 folds worth of models and than use those 5 models to make the predictions.\n\nYou Kaggle notebook only needs the models - in the case of a pytorch that would be *.pth files. I create a kaggle data set to hold those and add it to my inference notebook as data. Again, the pytorch shared example shows the result - you just need to learn how to do the data add :)", "votes": null }, { "id": "3154530", "postDate": "03/20/2025 03:17:23", "content": "Okay, thank you very much, because I can verify whether LeCun's normalization free approach is useful. Thank you again for your help
", "rawMarkdown": "Okay, thank you very much, because I can verify whether LeCun's normalization free approach is useful. Thank you again for your help", "votes": null }, { "id": "3155244", "postDate": "03/20/2025 21:05:12", "content": "You can technically submit but will only have 1 minute of runtime.>
\nI didn't understand the part about the 1-minute runtime. What does that mean?
", "rawMarkdown": "**You can technically submit but will only have 1 minute of runtime.>**\n\nI didn't understand the part about the 1-minute runtime. What does that mean?", "votes": null }, { "id": "3155352", "postDate": "03/21/2025 00:39:03", "content": "This is a problem when I submit. It seems that I should take out the trained model data and submission.csv and submit them separately using CPU resources.
", "rawMarkdown": "This is a problem when I submit. It seems that I should take out the trained model data and submission.csv and submit them separately using CPU resources.", "votes": null }, { "id": "3155444", "postDate": "03/21/2025 03:39:39", "content": "You have one minute if you use GPU by mistake for submission. Just another way of confirming that no GPU will be used for submission. It helps remind folks that they need to turn off accelerators.
", "rawMarkdown": "You have one minute if you use GPU by mistake for submission. Just another way of confirming that no GPU will be used for submission. It helps remind folks that they need to turn off accelerators.", "votes": null }, { "id": "3155533", "postDate": "03/21/2025 06:09:17", "content": "RuntimeError Traceback (most recent call last)
\n in ()
\n 109
\n 110 # Run Training and Inference
\n--> 111 train_model()
\n 112 run_inference()
in train_model()
\n 71 dummy_input = torch.randn((CONFIG[\"batch_size\"], 1, 128, CONFIG[\"target_length\"])).to(CONFIG[\"device\"])
\n 72 dummy_target = torch.randint(0, len(class_labels), (CONFIG[\"batch_size\"],)).to(CONFIG[\"device\"])
\n---> 73 outputs = model(dummy_input)
\n 74 loss = criterion(outputs, dummy_target)
\n 75 loss.backward()
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _wrapped_call_impl(self, *args, *kwargs)\n 1734 return self._compiled_call_impl(args, *kwargs) # type: ignore[misc]\n 1735 else:\n-> 1736 return self._call_impl(args, **kwargs)
\n 1737
\n 1738 # torchrec tests the code consistency with the following code
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _call_impl(self, *args, *kwargs)\n 1745 or _global_backward_pre_hooks or _global_backward_hooks\n 1746 or _global_forward_hooks or _global_forward_pre_hooks):\n-> 1747 return forward_call(args, **kwargs)
\n 1748
\n 1749 result = None
in forward(self, x)
\n 56 self.model = timm.create_model(\"efficientnet_b0\", pretrained=True, num_classes=num_classes)
\n 57 def forward(self, x):
\n---> 58 return self.model(x)
\n 59
\n 60 # 📌 Step 6: Train the Model
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _wrapped_call_impl(self, *args, *kwargs)\n 1734 return self._compiled_call_impl(args, *kwargs) # type: ignore[misc]\n 1735 else:\n-> 1736 return self._call_impl(args, **kwargs)
\n 1737
\n 1738 # torchrec tests the code consistency with the following code
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _call_impl(self, *args, *kwargs)\n 1745 or _global_backward_pre_hooks or _global_backward_hooks\n 1746 or _global_forward_hooks or _global_forward_pre_hooks):\n-> 1747 return forward_call(args, **kwargs)
\n 1748
\n 1749 result = None
/usr/local/lib/python3.10/dist-packages/timm/models/efficientnet.py in forward(self, x)
\n 267
\n 268 def forward(self, x):
\n--> 269 x = self.forward_features(x)
\n 270 x = self.forward_head(x)
\n 271 return x
/usr/local/lib/python3.10/dist-packages/timm/models/efficientnet.py in forward_features(self, x)
\n 250
\n 251 def forward_features(self, x):
\n--> 252 x = self.conv_stem(x)
\n 253 x = self.bn1(x)
\n 254 if self.grad_checkpointing and not torch.jit.is_scripting():
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _wrapped_call_impl(self, *args, *kwargs)\n 1734 return self._compiled_call_impl(args, *kwargs) # type: ignore[misc]\n 1735 else:\n-> 1736 return self._call_impl(args, **kwargs)
\n 1737
\n 1738 # torchrec tests the code consistency with the following code
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _call_impl(self, *args, *kwargs)\n 1745 or _global_backward_pre_hooks or _global_backward_hooks\n 1746 or _global_forward_hooks or _global_forward_pre_hooks):\n-> 1747 return forward_call(args, **kwargs)
\n 1748
\n 1749 result = None
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/conv.py in forward(self, input)
\n 552
\n 553 def forward(self, input: Tensor) -> Tensor:
\n--> 554 return self._conv_forward(input, self.weight, self.bias)
\n 555
\n 556
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/conv.py in _conv_forward(self, input, weight, bias)
\n 547 self.groups,
\n 548 )
\n--> 549 return F.conv2d(
\n 550 input, weight, bias, self.stride, self.padding, self.dilation, self.groups
\n 551 )
RuntimeError: Given groups=1, weight of size [32, 3, 3, 3], expected input[24, 1, 128, 256] to have 3 channels, but got 1 channels instead
", "rawMarkdown": "RuntimeError Traceback (most recent call last)\nThis looks like your input is a single channel, but your model requires three channels of input
", "rawMarkdown": "This looks like your input is a single channel, but your model requires three channels of input", "votes": null }, { "id": "3155543", "postDate": "03/21/2025 06:17:07", "content": "Yes, but I think it would be great if there were a better process for beginners. I'm curious why the test is not publicly available
", "rawMarkdown": "Yes, but I think it would be great if there were a better process for beginners. I'm curious why the test is not publicly available", "votes": null }, { "id": "3155665", "postDate": "03/21/2025 09:22:12", "content": "thx!!! im commit, but my score very low, just 0.54, soo hard
", "rawMarkdown": "thx!!! im commit, but my score very low, just 0.54, soo hard", "votes": null } ], "comments": [ { "id": 3154484, "author_name": "pcjimmmy", "author_url": "", "post_date": "03/20/2025 02:28:58", "content": "Need a bit more info - the 120 suggests the notebook ran that many minutes, while the limit for the competition is 90. So you got to cut at least 30 minutes out.
\nNot sure I understand the rest of your issues. Share what you copied?
", "votes": null, "replies": [ { "id": 3154488, "author_name": "yangjam", "author_url": "", "post_date": "03/20/2025 02:32:17", "content": "Your Notebook's runtime of 9 minutes exceeds this competition's GPU max of 1 minute. This is the issue I encountered. I don't know how to solve it
", "votes": null, "replies": [ { "id": 3154498, "author_name": "pcjimmmy", "author_url": "", "post_date": "03/20/2025 02:42:42", "content": "For prediction on test the notebook not suppose to use GPU.
\n\n\nGPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.>
\n
So disable GPU as a start point. The Accelerator drop down will let you select None - right now this probably shows one of the GPU's setups that can be used.
\nSince I do most my work on home pc I like to use three notebooks. The first one builds the data set. This competition has lots of options for doing things with the data. The 2nd notebook trains, while the third is for prediction. The third is the only one that I will upload to kaggle and run. My models are saved locally, so I will create a kaggle data set to add them to the prediction notebook on kaggle.
", "votes": null, "replies": [ { "id": 3154501, "author_name": "pcjimmmy", "author_url": "", "post_date": "03/20/2025 02:45:29", "content": "
Does this mean that training is a file, while testing should be pulled out as a separate notebook and used on the CPU?
", "votes": null, "replies": [ { "id": 3154513, "author_name": "pcjimmmy", "author_url": "", "post_date": "03/20/2025 02:59:09", "content": "Follow the example of several shared notebooks. They will generally train in one notebook and save the models. That can be done using GPU for speed.
\nThan a second notebook used to predict and submit. It must not use GPU. You add the saved models as data - a little tricky for new folks ……
", "votes": null, "replies": [ { "id": 3154520, "author_name": "yangjam", "author_url": "", "post_date": "03/20/2025 03:05:10", "content": "oh I roughly understand what you mean. I need to first train the model and save it to my local computer, then open Jupyter and upload my trained model as a prediction. This process cannot use GPU, is right?
", "votes": null, "replies": [ { "id": 3154529, "author_name": "pcjimmmy", "author_url": "", "post_date": "03/20/2025 03:13:52", "content": "Think your almost saying it right. Train a model anywhere you want. Local machine, goggle colab, borrow Elon's 200K super compter, etc. Save the MODELS - look at the pytorch shared notebooks. They save 5 folds worth of models and than use those 5 models to make the predictions.
\nYou Kaggle notebook only needs the models - in the case of a pytorch that would be *.pth files. I create a kaggle data set to hold those and add it to my inference notebook as data. Again, the pytorch shared example shows the result - you just need to learn how to do the data add :)
", "votes": null, "replies": [ { "id": 3154530, "author_name": "yangjam", "author_url": "", "post_date": "03/20/2025 03:17:23", "content": "Okay, thank you very much, because I can verify whether LeCun's normalization free approach is useful. Thank you again for your help
", "votes": null, "replies": [] } ] } ] } ] } ] }, { "id": 3155244, "author_name": "hotsonhonet", "author_url": "", "post_date": "03/20/2025 21:05:12", "content": "You can technically submit but will only have 1 minute of runtime.>
\nI didn't understand the part about the 1-minute runtime. What does that mean?
", "votes": null, "replies": [ { "id": 3155352, "author_name": "yangjam", "author_url": "", "post_date": "03/21/2025 00:39:03", "content": "This is a problem when I submit. It seems that I should take out the trained model data and submission.csv and submit them separately using CPU resources.
", "votes": null, "replies": [] }, { "id": 3155444, "author_name": "pcjimmmy", "author_url": "", "post_date": "03/21/2025 03:39:39", "content": "You have one minute if you use GPU by mistake for submission. Just another way of confirming that no GPU will be used for submission. It helps remind folks that they need to turn off accelerators.
", "votes": null, "replies": [ { "id": 3155543, "author_name": "yangjam", "author_url": "", "post_date": "03/21/2025 06:17:07", "content": "Yes, but I think it would be great if there were a better process for beginners. I'm curious why the test is not publicly available
", "votes": null, "replies": [] }, { "id": 3155665, "author_name": "yangjam", "author_url": "", "post_date": "03/21/2025 09:22:12", "content": "thx!!! im commit, but my score very low, just 0.54, soo hard
", "votes": null, "replies": [] } ] } ] } ] } ] } ] }, { "id": 3155533, "author_name": "rabiul9137", "author_url": "", "post_date": "03/21/2025 06:09:17", "content": "RuntimeError Traceback (most recent call last)
\n in ()
\n 109
\n 110 # Run Training and Inference
\n--> 111 train_model()
\n 112 run_inference()
in train_model()
\n 71 dummy_input = torch.randn((CONFIG[\"batch_size\"], 1, 128, CONFIG[\"target_length\"])).to(CONFIG[\"device\"])
\n 72 dummy_target = torch.randint(0, len(class_labels), (CONFIG[\"batch_size\"],)).to(CONFIG[\"device\"])
\n---> 73 outputs = model(dummy_input)
\n 74 loss = criterion(outputs, dummy_target)
\n 75 loss.backward()
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _wrapped_call_impl(self, *args, *kwargs)\n 1734 return self._compiled_call_impl(args, *kwargs) # type: ignore[misc]\n 1735 else:\n-> 1736 return self._call_impl(args, **kwargs)
\n 1737
\n 1738 # torchrec tests the code consistency with the following code
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _call_impl(self, *args, *kwargs)\n 1745 or _global_backward_pre_hooks or _global_backward_hooks\n 1746 or _global_forward_hooks or _global_forward_pre_hooks):\n-> 1747 return forward_call(args, **kwargs)
\n 1748
\n 1749 result = None
in forward(self, x)
\n 56 self.model = timm.create_model(\"efficientnet_b0\", pretrained=True, num_classes=num_classes)
\n 57 def forward(self, x):
\n---> 58 return self.model(x)
\n 59
\n 60 # 📌 Step 6: Train the Model
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _wrapped_call_impl(self, *args, *kwargs)\n 1734 return self._compiled_call_impl(args, *kwargs) # type: ignore[misc]\n 1735 else:\n-> 1736 return self._call_impl(args, **kwargs)
\n 1737
\n 1738 # torchrec tests the code consistency with the following code
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _call_impl(self, *args, *kwargs)\n 1745 or _global_backward_pre_hooks or _global_backward_hooks\n 1746 or _global_forward_hooks or _global_forward_pre_hooks):\n-> 1747 return forward_call(args, **kwargs)
\n 1748
\n 1749 result = None
/usr/local/lib/python3.10/dist-packages/timm/models/efficientnet.py in forward(self, x)
\n 267
\n 268 def forward(self, x):
\n--> 269 x = self.forward_features(x)
\n 270 x = self.forward_head(x)
\n 271 return x
/usr/local/lib/python3.10/dist-packages/timm/models/efficientnet.py in forward_features(self, x)
\n 250
\n 251 def forward_features(self, x):
\n--> 252 x = self.conv_stem(x)
\n 253 x = self.bn1(x)
\n 254 if self.grad_checkpointing and not torch.jit.is_scripting():
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _wrapped_call_impl(self, *args, *kwargs)\n 1734 return self._compiled_call_impl(args, *kwargs) # type: ignore[misc]\n 1735 else:\n-> 1736 return self._call_impl(args, **kwargs)
\n 1737
\n 1738 # torchrec tests the code consistency with the following code
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in _call_impl(self, *args, *kwargs)\n 1745 or _global_backward_pre_hooks or _global_backward_hooks\n 1746 or _global_forward_hooks or _global_forward_pre_hooks):\n-> 1747 return forward_call(args, **kwargs)
\n 1748
\n 1749 result = None
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/conv.py in forward(self, input)
\n 552
\n 553 def forward(self, input: Tensor) -> Tensor:
\n--> 554 return self._conv_forward(input, self.weight, self.bias)
\n 555
\n 556
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/conv.py in _conv_forward(self, input, weight, bias)
\n 547 self.groups,
\n 548 )
\n--> 549 return F.conv2d(
\n 550 input, weight, bias, self.stride, self.padding, self.dilation, self.groups
\n 551 )
RuntimeError: Given groups=1, weight of size [32, 3, 3, 3], expected input[24, 1, 128, 256] to have 3 channels, but got 1 channels instead
", "votes": null, "replies": [ { "id": 3155542, "author_name": "yangjam", "author_url": "", "post_date": "03/21/2025 06:15:47", "content": "This looks like your input is a single channel, but your model requires three channels of input
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3154459": "I am a beginner in Kaggle competitions and did not see any test datasets. Therefore, I copied a sample submission but was unable to upload it. The system informed me that 120 has exceeded the limit of 90", "3154484": "Need a bit more info - the 120 suggests the notebook ran that many minutes, while the limit for the competition is 90. So you got to cut at least 30 minutes out.\n\nNot sure I understand the rest of your issues. Share what you copied?", "3154488": "Your Notebook's runtime of 9 minutes exceeds this competition's GPU max of 1 minute. This is the issue I encountered. I don't know how to solve it", "3154498": "For prediction on test the notebook not suppose to use GPU.\n\n>GPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.>\n\nSo disable GPU as a start point. The Accelerator drop down will let you select None - right now this probably shows one of the GPU's setups that can be used.\n\nSince I do most my work on home pc I like to use three notebooks. The first one builds the data set. This competition has lots of options for doing things with the data. The 2nd notebook trains, while the third is for prediction. The third is the only one that I will upload to kaggle and run. My models are saved locally, so I will create a kaggle data set to add them to the prediction notebook on kaggle.", "3154501": "", "3154509": "Does this mean that training is a file, while testing should be pulled out as a separate notebook and used on the CPU?", "3154513": "Follow the example of several shared notebooks. They will generally train in one notebook and save the models. That can be done using GPU for speed. \n\nThan a second notebook used to predict and submit. It must not use GPU. You add the saved models as data - a little tricky for new folks ......", "3154520": "oh I roughly understand what you mean. I need to first train the model and save it to my local computer, then open Jupyter and upload my trained model as a prediction. This process cannot use GPU, is right?", "3154529": "Think your almost saying it right. Train a model anywhere you want. Local machine, goggle colab, borrow Elon's 200K super compter, etc. Save the MODELS - look at the pytorch shared notebooks. They save 5 folds worth of models and than use those 5 models to make the predictions.\n\nYou Kaggle notebook only needs the models - in the case of a pytorch that would be *.pth files. I create a kaggle data set to hold those and add it to my inference notebook as data. Again, the pytorch shared example shows the result - you just need to learn how to do the data add :)", "3154530": "Okay, thank you very much, because I can verify whether LeCun's normalization free approach is useful. Thank you again for your help", "3155244": "**You can technically submit but will only have 1 minute of runtime.>**\n\nI didn't understand the part about the 1-minute runtime. What does that mean?", "3155352": "This is a problem when I submit. It seems that I should take out the trained model data and submission.csv and submit them separately using CPU resources.", "3155444": "You have one minute if you use GPU by mistake for submission. Just another way of confirming that no GPU will be used for submission. It helps remind folks that they need to turn off accelerators.", "3155533": "RuntimeError Traceback (most recent call last)\n