huggingface / transformers

🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
https://huggingface.co/transformers
Apache License 2.0
133.73k stars 26.72k forks source link

Model Hub: Search by model size #10404

Closed ioana-blue closed 3 years ago

ioana-blue commented 3 years ago

🚀 Feature request

It would be great if the model cards for models would include the model size (i.e., the number of parameters) and then the model hub will allow searching for models by size.

Motivation

Depending on the task/problem/context, smaller or larger models are more beneficial. It's hard to keep up with all the models out there. For example, if I'm interested in distilled/compressed/smaller BERTs, I may be able to remember DistilBERT, MobileBERT but maybe forget about SqueezeBERT, TinyBERT, etc. A search by size would make all these smaller models visible.

julien-c commented 3 years ago

Definitely a good idea

ioana-blue commented 3 years ago

And since I started talking about model cards... :) I think it would be cool if you guys actually imposed some format. I think the original paper/idea had a format. Now "model cards" stands for "whatever the researcher had time to fill in that day" :) A few fields of interest: model size, training data, NLP tasks, language(s), paper, maybe something about model of inspiration (e.g., TinyBERT is a modification of BERT by...).

NielsRogge commented 3 years ago

I agree, there should at least be a template in my opinion. I hate to find models on the hub which don't provide any information. Moreover, all model cards look different, there's not really a structure.

julien-c commented 3 years ago

There is a template we link to in the second question of https://huggingface.co/docs (=> https://github.com/huggingface/model_card), though we should make it more built-in/central at some point.

david-waterworth commented 3 years ago

It would also be nice if the template also included details on tokenisation, what algorithm was used (BPE, Unigram, Word Piece) and the parameters (vocab size etc).

github-actions[bot] commented 3 years ago

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.