BAAI/BGE Reranker v2 M3
Lightweight 0.6B parameter cross-encoder built on bge-m3 foundation with LoRA fine-tuning and flash attention optimization. Strong multilingual support with fast inference, trained on diverse datasets including FEVER and MIRACL for production deployment efficiency.
Model Information
- Provider
- BAAI
- License
- Open Source
- Price per 1M tokens
- $0.020
- Release Date
- 2023-09-15
- Model Name
- bge-reranker-v2-m3
- Total Evaluations
- 2100
Performance Record
Wins693 (33.0%)
Losses1340 (63.8%)
Ties67 (3.2%)
Wins
Losses
Ties
Performance Overview
ELO ratings by dataset
BAAI/BGE Reranker v2 M3's ELO performance varies across different benchmark datasets, showing its strengths in specific domains.
BAAI/BGE Reranker v2 M3 - ELO by Dataset
Detailed Metrics
Dataset breakdown
Performance metrics across different benchmark datasets, including accuracy and latency percentiles.
DBPedia
ELO 146630.3% WR106W-242L-2T
Accuracy Metrics
- nDCG@5
- 0.715
- nDCG@10
- 0.778
- Recall@5
- 0.063
- Recall@10
- 0.106
Latency Distribution
- Mean
- 1332ms
- P50 (Median)
- 831ms
- P90
- 1455ms
business reports
ELO 142736.6% WR128W-215L-7T
Latency Distribution
- Mean
- 1143ms
- P50 (Median)
- 1106ms
- P90
- 1641ms
Scifact
ELO 142336.9% WR129W-221L-0T
Accuracy Metrics
- nDCG@5
- 0.843
- nDCG@10
- 0.860
- Recall@5
- 0.863
- Recall@10
- 0.906
Latency Distribution
- Mean
- 2928ms
- P50 (Median)
- 1581ms
- P90
- 1862ms
MSMARCO
ELO 139626.9% WR94W-204L-52T
Accuracy Metrics
- nDCG@5
- 0.985
- nDCG@10
- 0.985
- Recall@5
- 1.000
- Recall@10
- 1.000
Latency Distribution
- Mean
- 2176ms
- P50 (Median)
- 812ms
- P90
- 980ms
PG
ELO 139333.7% WR118W-232L-0T
Latency Distribution
- Mean
- 2457ms
- P50 (Median)
- 1019ms
- P90
- 1469ms
FiQa
ELO 136233.7% WR118W-226L-6T
Accuracy Metrics
- nDCG@5
- 0.112
- nDCG@10
- 0.120
- Recall@5
- 0.105
- Recall@10
- 0.130
Latency Distribution
- Mean
- 1309ms
- P50 (Median)
- 1316ms
- P90
- 1744ms
Compare Models
See how it stacks up
Compare BAAI/BGE Reranker v2 M3 with other top rerankers to understand the differences in performance, accuracy, and latency.