shibing624/chinese-text-correction-1.5b

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 12, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

The shibing624/chinese-text-correction-1.5b is a 1.5 billion parameter Chinese text correction model developed by shibing624, built upon the Qwen/Qwen2.5-1.5B-Instruct base model. This model is specifically fine-tuned for Chinese spelling and grammar correction, capable of handling both length-aligned errors like phonetic/visual similarities and grammatical mistakes, as well as length-unaligned errors such as missing or extra characters. It excels at comprehensive Chinese text correction tasks, offering a specialized solution for improving text accuracy.

Loading preview...

Overview

The shibing624/chinese-text-correction-1.5b model is a specialized 1.5 billion parameter language model, fine-tuned from Qwen/Qwen2.5-1.5B-Instruct, designed for comprehensive Chinese text correction. It addresses both Chinese Spelling Correction (CSC), which includes phonetic, visual, and grammatical errors that maintain text length, and Chinese Text Correction (CTC), which extends to correcting length-unaligned errors like missing or extra characters.

Key Capabilities

  • Spelling Correction: Identifies and corrects common Chinese spelling mistakes.
  • Grammar Correction: Rectifies grammatical inaccuracies in Chinese text.
  • Length-Aligned Error Handling: Proficient in correcting errors where the corrected text maintains the original length (e.g., 因该 to 应该).
  • Length-Unaligned Error Handling: Capable of addressing errors involving character additions or deletions, which change the text length.
  • Performance: Achieves an average F1 score of 0.6802 across various benchmarks, with notable scores of 0.7846 on EC-LAW and 0.9529 on MCSC datasets.

Good For

  • Automated Text Quality Improvement: Ideal for applications requiring high accuracy in Chinese text, such as content creation, educational tools, or customer service platforms.
  • Integration with pycorrector: Seamlessly integrates with the pycorrector library for easy deployment and usage in Python projects.
  • Developers needing a specialized Chinese correction model: Offers a focused solution compared to general-purpose LLMs, optimized for the nuances of Chinese text errors.