Slite/

ms-marco-MiniLM-L-4-v2-onnx-o4

from pathlib import Path
from transformers import AutoTokenizer
from optimum.onnxruntime import ORTModelForSequenceClassification, ORTOptimizer
from optimum.onnxruntime import AutoOptimizationConfig

model = "cross-encoder/ms-marco-MiniLM-L-4-v2"
tokenizer = AutoTokenizer.from_pretrained(model)
ort_model = ORTModelForSequenceClassification.from_pretrained(model, export=True)

save_dir = Path("/tmp/optimized_models")
save_dir.mkdir(exist_ok=True, parents=True)

optimizer = ORTOptimizer.from_pretrained(ort_model)
optimizer.optimize(
    optimization_config=AutoOptimizationConfig.O4(),
    save_dir=save_dir,
)

import torch
from transformers import AutoTokenizer
from transformers.pipelines.text_classification import ClassificationFunction
from optimum.pipelines import pipeline as ort_pipeline
from optimum.onnxruntime import ORTModelForSequenceClassification


model = "cross-encoder/ms-marco-MiniLM-L-4-v2"
device = torch.device(0) if torch.cuda.is_available() else -1
tokenizer = AutoTokenizer.from_pretrained(model)
ort_model = ORTModelForSequenceClassification.from_pretrained(
    model, file_name="model_optimized.onnx")
cross_encoder = ort_pipeline(
    task="text-classification",
    model=ort_model,
    tokenizer=tokenizer,
    device=device,
    function_to_apply=ClassificationFunction.SIGMOID,
    padding=True,
    truncation=True)

cross_encoder([{
    "text":
    "What is the purpose of life?",
    "text_pair":
    "The purpose of life is subjective and determined by each individual. Some may believe the purpose of life is to seek knowledge and education, to find happiness and fulfillment, or to live with purpose by helping others."
}])