1import torch
2import time
3from transformers import AutoTokenizer, pipeline
4from optimum.onnxruntime import ORTModelForSeq2SeqLM
5
6model = ORTModelForSeq2SeqLM.from_pretrained("chi-vi/hirashiba-mt-tiny-zh-vi-onnx")
7tokenizer = AutoTokenizer.from_pretrained("chi-vi/hirashiba-mt-tiny-zh-vi-onnx")
8
9with open("test.txt") as f:
10 text = f.readlines()
11
12start_time = time.time()
13
14pipe = pipeline("translation", model=model, tokenizer=tokenizer, device=0)
15result = pipe(text)
16print(result)
17
18print("--- %s seconds ---" % (time.time() - start_time))