if the community trains an adapter, where should they upload them?

... visually in the sub navigation

The linked SST adapter points to https://adapterhub.ml/adapters/ukp/bert-base-uncased-sst_pfeiffer/ which does not exist.

E.g., to explain the nd and wd in http://localhost:5000/adapters/pfeiffer/bert-base-multilingual-cased-cdo-wiki_pfeiffer/

https://arxiv.org/abs/2007.07779

Current Search works with substrings. E.g. "English NLI" does not retrieve any results. Implement a BoW approach to solve this issue

Some parts of the page are not yet mobile friendly, e.g., the header in the index page.

Tidy up the explore page and make the submenu / breadcrumbs user friendly. Add search functionality in JS for filtering

• JS filtering
• Task card style
• Task header style
• Adapter explorepage header style
• New adapter card style

• Add a new info page for each of the most relevant adapter architectures showing the config dict, description, citation and optionally an architecture image.
• Add links to this page from relevant places

Required updates for quickstart training & inference notebooks linked in the homepage:

• use stable version of adapter-transformers from PyPI
• avoid cloning HF transformers (confusing for users)
• switch from HF transformers GLUE processing (deprecated) to HF datasets

e.g., https://adapterhub.ml/adapters/ukp/roberta-base-rte_pfeiffer/

It would be helpful to include the adapter name. Here we can easily guess its name but it might not be as straightforward for others

Currently, the website just renders the config dictionary of the yaml info card.
The code snippet on the adapter page is invalid too for this reason.

• Improve direct navigation between different tasks on the explore_adapters page, e.g. by including in breadcrumb or by an additional nav.

We store this information in the yaml, we can also show it on the page if the contributors included their github or twitter handle

Which adapter by whom is the best, currently there is no way to identify which adapter, if trained on the same data, is the best to use

add all the adapters available

• Task Adapters

• Language Adapters

• include in schema
• display on website

Adapters need to be trained slightly different to fully fine-tuning. Add some info boxes for the insights we have gained while working with adapters

currently only the paper which has trained the adapter is referenced. We should also add this for the dataset which is used

Add the language of a dataset and display it on the task overview page (and basically everywhere where we include the task name).

@calpt I guess it needs a new field in the task yaml files?

extracts all the meta data available and stores it in a yaml file, ready for upload

AdapterHub - Adapter-Transformers v3 - Unifying Efficient Fine-Tuning

With the release of version 3.0 of adapter-transformers today, we're taking the first steps at integrating the grown and diversified landscape of efficient fine-tuning methods. Version 3.0 adds support for a first batch of recently proposed methods, including Prefix Tuning, Parallel adapters, Mix-and-Match adapters and Compacters. Further, improvements and changes to various aspects of the library are introduced.

https://adapterhub.ml/blog/2022/03/adapter-transformers-v3-unifying-efficient-fine-tuning/

extend the readme in repo to include information about adapter-transformers
rebrand dir to this new framework so that there are no conflicts with transformers

• Redirect upload page
• styling etc.?

• Add a new blog page using plain text files as posts

The spacing is currently not consistent. See image attached

They are now accepted and the bibtex should be updated accordingly (Hub and Website)

AdapterHub - Version 2 of AdapterHub Released

Today, we are releasing version 2 of the AdapterHub. This release introduces several exciting new ways for composing adapters through composition blocks, including AdapterFusion, parallel inference, Adapter stacking, and combinations thereof. Furthermore, we now support new Transformer architectures such as GPT-2 and BART.

https://adapterhub.ml/blog/2021/04/version-2-of-adapterhub-released/

If you search for e.g. "Natural Language Inference", nothing is shown

adapters need to be zipped separately, otherwise adapterhub doesnt find the files