WhereIsAI/UAE-Large-V1

Universal AnglE Embedding

📢 WhereIsAI/UAE-Large-V1 is licensed under MIT. Feel free to use it in any scenario. If you use it for academic papers, you could cite us via 👉 citation info.

🤝 Follow us on:

• GitHub: https://github.com/SeanLee97/AnglE.
• Arxiv: https://arxiv.org/abs/2309.12871 (ACL24)
• 📘 Document: https://angle.readthedocs.io/en/latest/index.html

Welcome to using AnglE to train and infer powerful sentence embeddings.

🏆 Achievements

• 📅 May 16, 2024 | AnglE's paper is accepted by ACL 2024 Main Conference
• 📅 Dec 4, 2024 | 🔥 Our universal English sentence embedding WhereIsAI/UAE-Large-V1 achieves SOTA on the MTEB Leaderboard with an average score of 64.64!

🧑‍🤝‍🧑 Siblings:

• WhereIsAI/UAE-Code-Large-V1: This model can be used for code or GitHub issue similarity measurement.

Usage

1. angle_emb

python -m pip install -U angle-emb

1. Non-Retrieval Tasks

There is no need to specify any prompts.

from angle_emb import AnglE
from angle_emb.utils import cosine_similarity

angle = AnglE.from_pretrained('WhereIsAI/UAE-Large-V1', pooling_strategy='cls').cuda()
doc_vecs = angle.encode([
    'The weather is great!',
    'The weather is very good!',
    'i am going to bed'
], normalize_embedding=True)

for i, dv1 in enumerate(doc_vecs):
    for dv2 in doc_vecs[i+1:]:
        print(cosine_similarity(dv1, dv2))

2. Retrieval Tasks

For retrieval purposes, please use the prompt Prompts.C for query (not for document).

from angle_emb import AnglE, Prompts
from angle_emb.utils import cosine_similarity

angle = AnglE.from_pretrained('WhereIsAI/UAE-Large-V1', pooling_strategy='cls').cuda()
qv = angle.encode(Prompts.C.format(text='what is the weather?'))
doc_vecs = angle.encode([
    'The weather is great!',
    'it is rainy today.',
    'i am going to bed'
])

for dv in doc_vecs:
    print(cosine_similarity(qv[0], dv))

2. sentence transformer

from angle_emb import Prompts
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("WhereIsAI/UAE-Large-V1").cuda()

qv = model.encode(Prompts.C.format(text='what is the weather?'))
doc_vecs = model.encode([
    'The weather is great!',
    'it is rainy today.',
    'i am going to bed'
])

for dv in doc_vecs:
    print(1 - spatial.distance.cosine(qv, dv))

Citation

If you use our pre-trained models, welcome to support us by citing our work:

@article{li2023angle,
  title={AnglE-optimized Text Embeddings},
  author={Li, Xianming and Li, Jing},
  journal={arXiv preprint arXiv:2309.12871},
  year={2023}
}