Hi! I have two questions if around selecting a new model to implement if you don’t mind.
Should I look to get confirmation from a Huggingface team member that a model is wanted before I start working on adding it, and should I check that a model meets a certain performance thresholds before asking? For example, there is a model mentioned in the issues that appears interesting but when run on examples it doesn’t seem to provide output that users might be interested in.
Does Huggingface have a list maintained of what new models they would like, or is the best place to look still this board and any open, unassigned issue with the ‘new model’ tag?
Some of the new models mentioned in GitHub issues appear to have interest, but it’s not explicit and it’s not clear how much interest there is.
On the docs around adding a new model what it says is, “To get started, open a New model addition issue for the model you want to see in Transformers. If you’re not especially picky about contributing a specific model, you can filter by the New model label to see if there are any unclaimed model requests and work on it”
Also, Sylvain mentioned here that support for models without requiring they be implemented in transformers will be more and more the case going forward here so that the library doesn’t grow exponentially.
Apologies if this has been asked before! I searched the forums and issues but didn’t find an answer. If helpful, I’d also be happy to add the context of the answer to the docs where relevant.