CoolFace
Modelpublic

iryneko571/mt5-translation-ja_zh-game-small

sourceHugging Facemitupdated 2y agoView on Hugging Face
0likes107downloads
Model Card

new model:

iryneko571/mt5-small-translation-ja_zh<br> better in most aspects, more like a base model with pure data<br> 数值上更好，是用更纯的数据跑的<br> includes colab notebook<br> 已经配置了colab的notebook，可以直接测试翻译，不需要安装<br>

Release Notes

  • —this model is finetuned from mt5-small
  • —will use about 1.5G vram, fp16 will be less than 1G(if batch size is small), cpu inference speed is ok anyway
  • —used a trimmed piece of pontoon dataset that features ja to zh translate part
  • —also scrambled bunch of the translation from mt5-translation-ja_zh-game-v0.1, which is a large amount of junk for training
  • —reason for making this model<br> Testing the ideas of using pontoon dataset<br> Constructing a flexible translation evaluation standard, need a poor performance model to compare

模型公开声明

  • —这个模型由 mt5-translation-ja_zh 继续训练得来
  • —使用大于1.5g的显存，fp16载入会小于1G显存（batch拉高会大于1G），使用cpu运作速度也还可以
  • —制作这个模型的原因<br> 尝试使用现有的模型精调，小模型训练速度奇快<br>
  • —本模型缺陷<br> 本身就是用来做测试的，虽然使用的显存很低，但翻译能力奇差<br>

简单的后端应用

还没稳定调试，慎用

  • —https://github.com/IryNeko/RabbitCafe

A more precise example using it

使用指南

python
from transformers import pipeline
model_name="iryneko571/mt5-translation-ja_zh-game-small"
#pipe = pipeline("translation",model=model_name,tokenizer=model_name,repetition_penalty=1.4,batch_size=1,max_length=256)
pipe = pipeline("translation",
  model=model_name,
  repetition_penalty=1.4,
  batch_size=1,
  max_length=256
  )

def translate_batch(batch, language='<-ja2zh->'): # batch is an array of string
    i=0 # quickly format the list
    while i<len(batch):
        batch[i]=f'{language} {batch[i]}'
        i+=1
    translated=pipe(batch)
    result=[]
    i=0
    while i<len(translated):
        result.append(translated[i]['translation_text'])
        i+=1
    return result

inputs=[]

print(translate_batch(inputs))

Roadmap

  • —Scamble more translation results from gpt4o, gpt3.5, claude, mt5 and other sources to make a more messy input
  • —increase translation accuracy
  • —apply lora on it and apply int8 inference to further decrease hardware requirements
  • —create onnx and ncnn model

how to find me

找到作者

Discord Server:<br> https://discord.gg/JmjPmJjA<br> If you need any help, a test server or just want to chat<br> 如果需要帮助，需要试试最新的版本，或者只是为了看下我是啥，可以进channel看看（这边允许发布这个吗？）<br>