Dzmitry Bahdanau

different number of queries in Academic, IMDB, etc. (compared to the original distribution)

11

Thank you for releasing this, great work! I have noticed that the "classical datasets" in your release always have less examples than in the original release. For example: Academic: 185...

generate_questions produces a dataset that is more balanced than the original one

4

There must be another discrepancy between `generate_questions.py` and the original script that was used to generate CLEVR. I have noticed that in CLEVR the answer distribution for counting questions is...

[The degeneracy check](https://github.com/facebookresearch/clevr-dataset-gen/blob/master/question_generation/generate_questions.py#L328 ) in `generate_questions.py` is applied only when there is a "raw" `relate` node in the template. This effectively means that it is only applied to question families...

original dataset and generation script output have different formats

3

In the dataset, "question_family_index" field takes values from 0 to 89. When I generate a new dataset with the generation script, "question_family_index" takes smaller values as it refers to the...

Your code performs a somewhat unusual L2 normalization across channels in the forward pass ([here](https://github.com/ronghanghu/n2nmn/blob/master/models_shapes/nmn3_modules.py#L60)), which is not mentioned in the paper, I think. Can you please share your motivation...

Workflow improvement suggestions

3

I really like using smart_dispatch and I have a few ideas how it could be improved: - The location for creating logs should configurable, now `SMART_DISPATCH_LOGS` seems to be hardcoded....

KeyError when inhering from a self-referencing generic model

3

### Initial Checks - [X] I confirm that I'm using Pydantic V2 ### Description I have this very short code snippet below that crashes with `KeyError`, see stack trace. Should...