Include a readable repr for ModelInfo

[muellerzr]

This includes the start of a better __repr__ for the ModelInfo class, started from this issue: #31

This is based on what I've started for the AdaptNLP

A few points of difference that could be done here is we could also programmatically include the tasks available for it, similar to what was done in AdaptNLP, but I guess the key information is what would be the best-case scenario for a model repr? Currently what I have lets it look like so:

Model Name: "gpt2", Tags: ['pytorch', 'tf', 't5'], Task: ['summarization']

For adapt I go a bit further (with the whole programmatic way I mentioned early) by checking what tags are also tasks and assigning them then.

The last thing is if we want something like:

Model Name: "gpt2-hybrid", Author: "username", Tags: ..., Task: ...

What are your thoughts?

cc: @LysandreJik

[LysandreJik]

I think the proposition of making repr more understandable is great! Thinking about from a high-level perspective, I understand that the difference between __repr__ and __str__ is that the former is for developers (useful when debugging), should be unambiguous, while the latter is for end-users and supposed to be readable.

To that end, I'm wondering if it wouldn't be better to have a __repr__ that shows a bit more information, like the __dict__, while the __str__ outputs the current __repr__ output. What do you think?

[muellerzr]

@LysandreJik I do like that idea, let me work on that this weekend and I'll get back to you with what I come up with in regards to a presentable format/strings

[muellerzr]

@LysandreJik apologies it took so long, had some final exams and projects that needed to get done!

Here's my adjustments I've made, the __repr__ now gives an output like so:

ModelInfo: {
	modelId: distilgpt2
	sha: None
	tags: ['pytorch', 'tf', 'tflite', 'rust', 'gpt2', 'lm-head', 'causal-lm', 'en', 'dataset:openwebtext', 'exbert', 'license:apache-2.0', 'text-generation']
	pipeline_tag: text-generation
	siblings: [<huggingface_hub.hf_api.ModelFile object at 0x7efdbba60090>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba600d0>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba60110>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba60150>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba60190>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba60210>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba60250>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba60290>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba602d0>, <huggingface_hub.hf_api.ModelFile object at 0x7efdbba601d0>]
	private: False
	key: 
}

Which I think is more aligned to what you were wanting, while the str has the original, IE:

Model Name: distilgpt2, Tags: ['pytorch', 'tf', 'tflite', 'rust', 'gpt2', 'lm-head', 'causal-lm', 'en', 'dataset:openwebtext', 'exbert', 'license:apache-2.0', 'text-generation'], Task: text-generation

Do let me know your thoughts please! :)

[LysandreJik]

Looks good to me! Thanks @muellerzr!

[LysandreJik]

Before we merge, you seem to have code quality issues, could you run the following from the root of your clone?

pip install .[quality]
make style
make quality

[muellerzr]

@LysandreJik it seems to have picked up an unrelated file with it, but I ran the code quality

[LysandreJik]

Fantastic, thanks a lot for working on this!