Transformers EncoderDecoder language model on Optimum OnnxRuntime

infer_onnx.py

# pip install transformers, optimum, onnx, onnxruntime, fugashi, unidic-lite
from transformers import BertJapaneseTokenizer,PreTrainedTokenizerFast
from optimum.onnxruntime import ORTModelForSeq2SeqLM

encoder_model_name = "cl-tohoku/bert-base-japanese-v2"
decoder_model_name = "skt/kogpt2-base-v2"

# using local tokenizer
# encoder_model_name = "./src_tokenizer"
# decoder_model_name = "./trg_tokenizer"

src_tokenizer = BertJapaneseTokenizer.from_pretrained(encoder_model_name)
trg_tokenizer = PreTrainedTokenizerFast.from_pretrained(decoder_model_name)

# `from_transformers=True` downloads the PyTorch weights and converts them to ONNX format
model = ORTModelForSeq2SeqLM.from_pretrained("./onnx")
text = "ギルガメッシュ討伐戦"
text2 = "ギルガメッシュ討伐戦に行ってきます。一緒に行きましょうか？"

def translate(text_src):
    embeddings = src_tokenizer(text_src, return_attention_mask=False, return_token_type_ids=False, return_tensors='pt')
    embeddings = {k: v for k, v in embeddings.items()}
    output = model.generate(**embeddings)[0, 1:-1]
    text_trg = trg_tokenizer.decode(output.cpu())
    return text_trg

print(translate(text))
print(translate(text2))

setup.py

from cx_Freeze import setup, Executable 
import sys 
sys.setrecursionlimit(5000)

buildOptions = {
	"packages":
    ["transformers","fugashi","unidic_lite","optimum"]
    ,
    "excludes":[
    	"matplotlib"
    ]
}
 
exe = [Executable("infer_onnx.py")]  # 2 
# 3 
setup(name= 'FFJaKo',
    version = '0.1',
    author = "sappho192",
    description = "FFXIV Ja-Ko Translator",
    options = dict(build_exe = buildOptions),
    executables = exe)