Teradata/multilingual-e5-base

1import onnxruntime as rt2 3from sentence_transformers.util import cos_sim4from sentence_transformers import SentenceTransformer5 6import transformers7 8import gc9import json10 11 12with open('conversion_config.json') as json_file:13    conversion_config = json.load(json_file)14 15 16    model_id = conversion_config["model_id"]17    number_of_generated_embeddings = conversion_config["number_of_generated_embeddings"]18    precision_to_filename_map = conversion_config["precision_to_filename_map"]19    20    sentences_1 = 'How is the weather today?'21    sentences_2 = 'What is the current weather like today?'22    23    print(f"Testing on cosine similiarity between sentences: \n'{sentences_1}'\n'{sentences_2}'\n\n\n")24    25    tokenizer = transformers.AutoTokenizer.from_pretrained("./")26    enc1 = tokenizer(sentences_1)27    enc2 = tokenizer(sentences_2)28    29    for precision, file_name in precision_to_filename_map.items():30    31        32        onnx_session = rt.InferenceSession(file_name)33        embeddings_1_onnx = onnx_session.run(None,     {"input_ids": [enc1.input_ids], 34             "attention_mask": [enc1.attention_mask]})[1][0]35    36        embeddings_2_onnx = onnx_session.run(None,     {"input_ids": [enc2.input_ids], 37             "attention_mask": [enc2.attention_mask]})[1][0]38    39        del onnx_session40        gc.collect()41        print(f'Cosine similiarity for ONNX model with precision "{precision}" is {str(cos_sim(embeddings_1_onnx, embeddings_2_onnx))}')42    43    44    45    46    model = SentenceTransformer(model_id, trust_remote_code=True)47    embeddings_1_sentence_transformer = model.encode(sentences_1, normalize_embeddings=True, trust_remote_code=True)48    embeddings_2_sentence_transformer = model.encode(sentences_2, normalize_embeddings=True, trust_remote_code=True)49    print('Cosine similiarity for original sentence transformer model is '+str(cos_sim(embeddings_1_sentence_transformer, embeddings_2_sentence_transformer)))