nlpai-lab / feature-extraction Open weights updated 1 month ago

KURE-v1

Name: KURE-v1
Author: nlpai-lab

Introducing Korea University Retrieval Embedding model, KURE-v1 It has shown remarkable performance in Korean text retrieval, speficially overwhelming most multilingual embedding models. To our knowledge, It is one of the best publicly opened Korean retrieval models.

Params

568 M

Context

8,192

Downloads 30d

578 K

Likes

100

Commercial use: allowed mit Not gated SAFETENSORS View on Hugging Face ↗

Download history

daily snapshots · 22 days

663 K400 K

Oct 3Oct 10Oct 17Oct 24

Can you run it?

Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.

File	Quant	Size	Est. VRAM	Verdict on RTX 4090 · 24 GB
model.safetensors	f32	2.3 GB	3.1 GB	✅ Runs comfortably

Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.

Specifications

Architecture: XLMRobertaModel
Parameters: 568 M
Tensor type: F32
Context length: 8,192
Vocabulary: 250,002
Layers / heads: 24 / 16
Licence: mit
First seen on the Hub: 2024-12-18
Base model: bge-m3
Training datasets: undisclosed
Added to our catalog: 2026-10-03

Family

Base model and the most-downloaded derivatives in the catalog.

Base bge-m3 BAAI 29.6 M

Compare with any feature-extraction model

Popular comparisons

bge-small-en-v1.5 54.0 M bge-large-en-v1.5 9.7 M bge-base-en-v1.5 8.6 M Qwen3-Embedding-0.6B 8.1 M multilingual-e5-large 6.5 M bge-reranker-large 1.9 M