Initialize Basic Organization #1
Conversation
|
Hi @abheesht17 @blackrosedragon2, There is a bit of an issue with configs. While the latest omegaconf (2.1 dev), we can use the YAML Merge Syntax (See this PR, however, it is not compatible with Variable Interpolation (See this Issue). While, we can keep the configs to either have merges, or variable interpolation, but not both, we need to find better alternatives of configuration management to allow both. A clean way would be either
Let me know which one is preferred, I'll change things accordingly. |
|
Also, the code for the dataset class has some issues, which need to be addressed. |
… custom training loop
… custom training loop
| ReLU(), | ||
| MaxPool2d(kernel_size=2, stride=2), | ||
| ) | ||
| self.linear_layers = Linear(4 * 7 * 7, 10) | ||
| self.linear_layers = Linear(32 * 3 * 3, 10) | ||
|
|
||
| def forward(self, x_in): | ||
| x_out = self.cnn_layers(x_in["image"]) |
There was a problem hiding this comment.
We can add an element in the model config, huggingface_bool. If True, we can return the loss as well from the forward function.
There was a problem hiding this comment.
The labels are there already.
gchhablani
changed the title
Add initial code structure for running experiments. I will be adding a simple CNN model, with MNIST dataset and test it on Colab and then make it ready for review.
Once done, it should be merged by the reviewers using "Squash and merge" option. This will help in cleaner commits.
Also, even small changes should be made using PR, and after at least one of reviewers reviews them.
Currently, I am not writing the code for GridSearch as we have switched to OmegaConf configurations, I'll add that in a separate PR.