bge-reranker-v2-m3

Multi-format version of BAAI/bge-reranker-v2-m3 - optimized for deployment.

Model Information

Property	Value
Base Model	BAAI/bge-reranker-v2-m3
Task	text-classification
Type	Text Model
Trust Remote Code	False

Available Versions

Folder	Format	Description	Size
safetensors-fp32/	PyTorch FP32	Baseline, highest accuracy	2187 MB
safetensors-fp16/	PyTorch FP16	GPU inference, ~50% smaller	1104 MB

Usage

PyTorch (GPU)

from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

# GPU inference with FP16
model = AutoModelForSequenceClassification.from_pretrained(
    "n24q02m/bge-reranker-v2-m3",
    subfolder="safetensors-fp16",
    torch_dtype=torch.float16
).cuda()
tokenizer = AutoTokenizer.from_pretrained(
    "n24q02m/bge-reranker-v2-m3",
    subfolder="safetensors-fp16"
)

# Rerank inference
pairs = [["what is panda?", "The giant panda is a bear species."]]
with torch.no_grad():
    inputs = tokenizer(pairs, padding=True, truncation=True, return_tensors="pt").to("cuda")
    scores = model(**inputs).logits.view(-1,).float()
print(scores)

Notes

1. SafeTensors FP16 is the primary format for GPU inference

License

Apache 2.0 (following the base model's license)

Credits

• Base Model: BAAI/bge-reranker-v2-m3
• Conversion: PyTorch + SafeTensors