Views
No views yet
"passage: " prefix for documents and "query: " prefix for search queries (e5 model convention).1from optimum.onnxruntime import ORTModelForFeatureExtraction
2from transformers import AutoTokenizer
3import numpy as np
4
5tokenizer = AutoTokenizer.from_pretrained("thomasbeste/multilingual-e5-large-onnx-int8")
6model = ORTModelForFeatureExtraction.from_pretrained("thomasbeste/multilingual-e5-large-onnx-int8")
7
8inputs = tokenizer("passage: Your text here", return_tensors="np", padding=True, truncation=True)
9outputs = model(**inputs)
10embedding = outputs.last_hidden_state.mean(axis=1) # Mean pooling
11embedding = embedding / np.linalg.norm(embedding) # L2 normalize