Downloads · 30 days
9
17% of all-time downloads
bge-reranker-v2-m3-onnx-fp16
LaraAI-Labs/bge-reranker-v2-m3-onnx-fp16
bge-reranker-v2-m3-onnx-fp16 is a machine learning model from LaraAI-Labs. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for onnx. The card lists the license as apache-2.0.
BAAI/bge-reranker-v2-m3 (XLM-RoBERTa-large cross-encoder) ONNX 변환본. 온디바이스 한국어 의미검색 리랭킹(retrieve→rerank)용.
All-time downloads
52
Public
Repo size
1.2 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.onnx1.1 GB · 98%
At a glance
- Library
- onnx
- License
- apache-2.0
- Model type
- xlm-roberta
- Access
- Public
- Created
- Jun 22, 2026
- Updated
- Jun 22, 2026
- SHA
- 9e896a97
Base models
Model card
From the Hugging Face model README
bge-reranker-v2-m3 ONNX (FP16, fp32 output)
BAAI/bge-reranker-v2-m3 (XLM-RoBERTa-large
cross-encoder) ONNX 변환본. 온디바이스 한국어 의미검색 리랭킹(retrieve→rerank)용.
스펙
- 정밀도: fp16 가중치 + fp32 출력(logit). CUDA EP 가속용(int8 은 CUDA MatMulInteger 미지원→느림). 소비측(C++ 리랭커) 출력 버퍼 fp32 계약 충족. ~1.1GB, opset 17.
- 입력:
input_ids,attention_mask(XLM-R 쌍 인코딩<s> query </s></s> passage </s>). 출력:logits[batch, 1]= 관련도(높을수록 관련). - 토크나이저: XLM-R SentencePiece(
sentencepiece.bpe.model동봉). - j1(Jetson AGX Orin) 실측: CUDA EP, 16쌍 배치 warm ~100ms.
생성 (재현)
python export_reranker_onnx.py --model BAAI/bge-reranker-v2-m3 --out ./reranker --dtype fp16
torch.onnx.export(eager) → 출력 .float() 캐스트. (int8 변형도 가능하나 GPU 가속엔 fp16.)
라이선스
원본 BAAI/bge-reranker-v2-m3 (Apache-2.0) 의 파생.
- Library
- onnx
- Type
- xlm-roberta
- License
- apache-2.0
- Languages
- ko, multilingual
- Created
- Jun 22, 2026
- Updated
- Jun 22, 2026