Uiversal AglE Embeddig

Follow us o:

• GitHub: https://github.com/SeaLee97/AglE.
• Arxiv: https://arxiv.org/abs/2309.12871

? Our uiversal Eglish setece embeddig WhereIsAI/UAE-Large-V1 achieves SOTA o the MTEB Leaderboard with a average score of 64.64!

Usage

pytho -m pip istall -U agle-emb

1) No-Retrieval Tasks

from agle_emb import AglE

agle = AglE.from_pretraied('WhereIsAI/UAE-Large-V1', poolig_strategy='cls').cuda()
vec = agle.ecode('hello world', to_umpy=True)
prit(vec)
vecs = agle.ecode(['hello world1', 'hello world2'], to_umpy=True)
prit(vecs)

2) Retrieval Tasks

For retrieval purposes, please use the prompt Prompts.C.

from agle_emb import AglE, Prompts

agle = AglE.from_pretraied('WhereIsAI/UAE-Large-V1', poolig_strategy='cls').cuda()
agle.set_prompt(prompt=Prompts.C)
vec = agle.ecode({'text': 'hello world'}, to_umpy=True)
prit(vec)
vecs = agle.ecode([{'text': 'hello world1'}, {'text': 'hello world2'}], to_umpy=True)
prit(vecs)

Citatio

If you use our pre-traied models, welcome to support us by citig our work:

@article{li2023agle,
  title={AglE-optimized Text Embeddigs},
  author={Li, Xiamig ad Li, Jig},
  joural={arXiv preprit arXiv:2309.12871},
  year={2023}
}

Universal AnglE Embedding Follow us on: GitHub: https://github.com/SeanLee97/AnglE. Arxiv: https://