📢 New preprint 🎉
We - the AdapterHub team - present the M2QA benchmark to evaluate joint domain and language transfer!
🔬 Key highlight: We show that adapter-based methods on small language models can reach the performance of Llama 3 on M2QA! 🚀
👇
🚀Wanna help us shape the future of the adapters library?🔎Take our survey!✏️
🧠Your input is crucial in planning our next steps. Share your thoughts in a 5 min. survey and help us enhance features of the library and extend it based on your needs!
--> https://t.co/5IcM7bcXso
12 new Adapters for models xlm-roberta-large, xlm-roberta-base, bert-base-multilingual-cased by @yyyyyyyyifan for tasks mlki/es, mlki/ts, mlki/tp and more!
Here's one to check out: https://t.co/kNhqyxl4bE
New Adapter by @https://twitter.com/kabirahuja004:
Pfeiffer adapter stacked on top of language adapter for the NLI task.
Check it out at https://t.co/28I33RR2vF
New Adapter by @kalpeshk2011:
This adapter has been trained on the English formality classification GYAFC dataset and tested with other language adapters (like hindi) for zero-shot transfer.
Check it out at https://t.co/LM4YJsgADE
New Adapter by @clifapt:
Adapter for mbart-large-cc25 in Pfeiffer architecture with reduction factor 2 trained on the WMT16 Romanian-English translation task.
Check it out at https://t.co/iWbSjcFxfp
22 new Adapters for model bert-base-multilingual-cased by @PfeiffJo for tasks wikiann/en, zh_yue/wiki, cs/wiki and more!
Here's one to check out: https://t.co/ColBlujA6i
16 new Adapters for model facebook/bart-base by @clifapt for tasks nli/qnli, lingaccept/cola, sts/mrpc and more!
Here's one to check out: https://t.co/oGIlJTlCQT
25 new Adapters for model distilbert-base-uncased by @clifapt for tasks sentiment/sst-2, rc/race, rc/multirc and more!
Here's one to check out: https://t.co/D61rRbIh81
Check out our paper “AdapterDrop"!
We find that Adapters can train 60% faster than full fine-tuning. With AdapterDrop we increase inference speed by up to 36% for 8 parallel tasks.
GGeigle @Maxxx216@devnull90@PfeiffJo NReimers IGurevych @AdapterHub
https://t.co/cXoKcCDBKK
In our new paper we show that randomly dropping out adapters during training results in a robust dynamically scalable model!
Also adapter weights can be shared across layers 😲
Check it out 👇
New Adapter by @NirantK, @meghanabhange:
Adapter for Hinglish Sentiment Analysis, based on SemEval 2020 Task 9.
Check it out at https://t.co/KjbQFlyEAc
New Adapter by @mapmeld:
Adapter for AraBERT (aubmindlab/bert-base-arabert) trained to classify Arabic by dialect, trained for 3 epochs on samples from University of British Columbia and John Hopkins University.
Check it out at https://t.co/TBMQ1ZZ5Iw