metadata

language:
  - zh
tags:
  - chatglm
  - pytorch
  - zh
  - Text2Text-Generation
license: apache-2.0
widget:
  - text: |-
      对下面中文拼写纠错：
      少先队员因该为老人让坐。
      答：

Chinese Spelling Correction LoRA Model

ChatGLM3-6B中文纠错LoRA模型

shibing624/chatglm3-6b-csc-chinese-lora evaluate test data：

The overall performance of shibing624/chatglm3-6b-csc-chinese-lora on CSC test:

input_text	pred
对下面文本纠错：少先队员因该为老人让坐。	少先队员应该为老人让座。

在CSC测试集上生成结果纠错准确率高，由于是基于THUDM/chatglm3-6b模型，结果常常能带给人惊喜，不仅能纠错，还带有句子润色和改写功能。

Usage

本项目开源在 pycorrector 项目：pycorrector，可支持ChatGLM原生模型和LoRA微调后的模型，通过如下命令调用：

Install package:

pip install -U pycorrector

from pycorrector import GptCorrector
model = GptModel("THUDM/chatglm3-6b", "chatglm", peft_name="shibing624/chatglm3-6b-csc-chinese-lora")
r = model.correct_batch(["少先队员因该为老人让坐。"])
print(r) # ['少先队员应该为老人让座。']

Usage (HuggingFace Transformers)

Without pycorrector, you can use the model like this:

First, you pass your input through the transformer model, then you get the generated sentence.

Install package:

pip install transformers

import os

import torch
from peft import PeftModel
from transformers import AutoTokenizer, AutoModel

os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm3-6b", trust_remote_code=True)
model = AutoModel.from_pretrained("THUDM/chatglm3-6b", trust_remote_code=True).half().cuda()
model = PeftModel.from_pretrained(model, "shibing624/chatglm3-6b-csc-chinese-lora")

sents = ['对下面文本纠错\n\n少先队员因该为老人让坐。',
         '对下面文本纠错\n\n下个星期，我跟我朋唷打算去法国玩儿。']


def get_prompt(user_query):
    vicuna_prompt = "A chat between a curious user and an artificial intelligence assistant. " \
                    "The assistant gives helpful, detailed, and polite answers to the user's questions. " \
                    "USER: {query} ASSISTANT:"
    return vicuna_prompt.format(query=user_query)


for s in sents:
    q = get_prompt(s)
    input_ids = tokenizer(q).input_ids
    generation_kwargs = dict(max_new_tokens=128, do_sample=True, temperature=0.8)
    outputs = model.generate(input_ids=torch.as_tensor([input_ids]).to('cuda:0'), **generation_kwargs)
    output_tensor = outputs[0][len(input_ids):]
    response = tokenizer.decode(output_tensor, skip_special_tokens=True)
    print(response)

output:

少先队员应该为老人让座。
下个星期，我跟我朋友打算去法国玩儿。

模型文件组成：

chatglm3-6b-csc-chinese-lora
    ├── adapter_config.json
    └── adapter_model.bin

训练参数：

num_epochs: 5
per_device_train_batch_size: 6
learning_rate: 2e-05
best steps: 25100
train_loss: 0.0834
lr_scheduler_type: linear
base model: THUDM/chatglm3-6b
warmup_steps: 50
"save_strategy": "steps"
"save_steps": 500
"save_total_limit": 10
"bf16": false
"fp16": true
"optim": "adamw_torch"
"ddp_find_unused_parameters": false
"gradient_checkpointing": true
max_seq_length: 512
max_length: 512
prompt_template_name: vicuna
6 * V100 32GB, training 48 hours

训练数据集

训练集包括以下数据：

中文拼写纠错数据集：https://huggingface.co/datasets/shibing624/CSC
中文语法纠错数据集：https://github.com/shibing624/pycorrector/tree/llm/examples/data/grammar
通用GPT4问答数据集：https://huggingface.co/datasets/shibing624/sharegpt_gpt4

如果需要训练文本纠错模型，请参考https://github.com/shibing624/pycorrector

Citation

@software{pycorrector,
  author = {Ming Xu},
  title = {pycorrector: Text Error Correction Tool},
  year = {2023},
  url = {https://github.com/shibing624/pycorrector},
}