ONNX GPU Runtime with O4 for BAAI/bge-reranker-large

benchmark: https://colab.research.google.com/drive/1HP9GQKdzYa6H9SJnAZoxJWq920gxwd2k

Convert

!optimum-cli export onnx -m BAAI/bge-reranker-large --optimize O4 bge-reranker-large-onnx-o4 --device cuda

Usage

# pip install "optimum[onnxruntime-gpu]" transformers

from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained('swulling/bge-reranker-large-onnx-o4')
model = ORTModelForSequenceClassification.from_pretrained('swulling/bge-reranker-large-onnx-o4')
model.to("cuda")

pairs = [['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']]
with torch.no_grad():
    inputs = tokenizer(pairs, padding=True, truncation=True, return_tensors='pt', max_length=512)
    scores = model(**inputs, return_dict=True).logits.view(-1, ).float()
    print(scores)

Source model

https://huggingface.co/BAAI/bge-reranker-large