Add CIFAR10 dataset, dataloader and training scripts #76
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds the code required to train a CNN on the CIFAR10 dataset.
I adapted the Dataloader already there to one that looks more like a Pytorch dataloader.
I would make it generic, but at this moment Mojo doesn't seem to support generic assignment?
I.e. I created a BaseDataset, so that the DataLoader can possibly call this, but I get a compiler error:
TODO: dynamic traits not supported yet, please use a compile time generic instead of 'BaseDataset'
Maybe one of you knows how to make the compile time generics work here?
I haven't fully checked how well the model trains (add accuracy?), but the loss is about the same for PyTorch and Basalt.
Could also maybe increase the model size for CIFAR 10.
The BasaltTensor
__init__(inout self, owned tensor: _Tensor[dtype]):
results in some kind of double free making the program segfault, so it's just a simple copy for the moment.