Zach Mueller
Zach Mueller
Should use `delegates()` to propagate `dls_kwargs` instead of just `dls_kwargs`
Interpretation grabs the inputs as part of `get_preds`, which for larger datasets/dataloaders this can cause out of memory issues. Solution would be to integrate in a look up for grabbing...
When decoding with `Categorize`, it will return a `Category`, such as: ```python learn.dls.categorize.decode(preds[2]) "(#3) ['Siamese','chihuahua','american_bulldog']" ``` Which is a string representation of a list, not quite what we want. Not...
A warning should be added to `MixHandler` for now that raises a `NotImplementedError` if anyone tries to do something other than `CategoryBlock`/Single label classification and use the callback (and related...
One nice feature would be if `get_preds` could return the class names rather than `tensor(0)` when passing `with_decoded`. I've done this within fastinference but of course this isn't nice as...
Ideally we should be able to simplify the logic of learn.summary by checking for shape changes rather than specific layers, and then depending on what comes of that we can...
## 🐛 Bug Seems that aim's UI interface for Colab seems to be broken right now 😢 ### To reproduce 1. `!pip install aim` 2. `%load_ext aim` 3. `%aim up`...
Currently the test suit takes 10+ minutes to run, depending on your computer. I ran them all with `--durations=0` so we can see how long each of them take. **Ideal...
Training API will use `fastai` under the hood, and we'll make a few functions to build general datasets. ## Tasks and sample datasets to use: - [x] Language Models: *...