XLNet-japanese

Model description

This model require Mecab and senetencepiece with XLNetTokenizer. See details https://qiita.com/mkt3/items/4d0ae36f3f212aee8002

This model uses NFKD as the normalization method for character encoding. Japanese muddle marks and semi-muddle marks will be lost.

日本語の濁点・半濁点がないモデルです

How to use

from fugashi import Tagger 

from transformers import (
    pipeline,
    XLNetLMHeadModel,
    XLNetTokenizer
)

class XLNet():
    def __init__(self):
        self.m = Tagger('-Owakati') 
        self.gen_model = XLNetLMHeadModel.from_pretrained("hajime9652/xlnet-japanese")
        self.gen_tokenizer = XLNetTokenizer.from_pretrained("hajime9652/xlnet-japanese")
         
    def generate(self, prompt="福岡のご飯は美味しい。コンパクトで暮らしやすい街。"):
        prompt = self.m.parse(prompt)
        inputs = self.gen_tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
        prompt_length = len(self.gen_tokenizer.decode(inputs[0], skip_special_tokens=True, clean_up_tokenization_spaces=True))
        outputs = self.gen_model.generate(inputs, max_length=200, do_sample=True, top_p=0.95, top_k=60)
        generated = prompt + self.gen_tokenizer.decode(outputs[0])[prompt_length:]
        return generated

Limitations and bias

This model's training use the Japanese Business News.

Important matter

The company that created and published this model is called Stockmark. This repository is for use by HuggingFace and not for infringement. See this documents https://qiita.com/mkt3/items/4d0ae36f3f212aee8002 published by https://github.com/mkt3

Downloads last month
510
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.