Cosmos BERT
Make it so that future adapters, finetunes can be trained to work with texts that are both in English and Japanese.
BERT serves as an early replacement for the T5/Qwen model.
It was trained on both T5 (text) and the AnimaTextToImagePipeline (text-image pairs).
Larger and smarter
The model is designed to accept input in both English and Japanese, and the token length is no longer limited to 77/248/512 tokens.
The fallback function, which processes English-only text, works out of the box. Japanese words are efficiently tokenized in any case.
- Downloads last month
- 36
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for nightknocker/cosmos-bert-japanese
Base model
sbintuitions/modernbert-ja-310m
Finetuned
retrieva-jp/amber-large