PeterAM4/deepseek-paraphrase
PeterAM4/deepseek-paraphrase is a 7.6 billion parameter language model fine-tuned from deepseek-ai/DeepSeek-R1-Distill-Qwen-7B, specialized for high-quality paraphrase generation. It was trained using LoRA on multi-domain text and the Quora paraphrase dataset, offering a 32768 token context length. This model excels at semantic preservation and lexical diversity, outperforming standard paraphrasing models like BART and T5, making it ideal for applications requiring robust text rephrasing.
Loading preview...
Deepseek Paraphrase Overview
PeterAM4/deepseek-paraphrase is a 7.6 billion parameter language model, fine-tuned from deepseek-ai/DeepSeek-R1-Distill-Qwen-7B, specifically for high-quality paraphrase generation. This model leverages LoRA (Low-Rank Adaptation) for efficient training and has been merged back into the base model for streamlined inference. It is designed to rephrase text while preserving its original meaning and varying its wording and structure.
Key Capabilities
- Specialized Paraphrasing: Fine-tuned on a diverse dataset including literary works, technical documentation, academic papers, and the Quora paraphrase dataset.
- Strong Performance: Outperforms standard paraphrasing models like BART and T5 in key metrics, demonstrating excellent semantic preservation (BERTScore: 0.952) and acceptable lexical diversity (BLEU Diversity: 0.513).
- Context Length: Supports a substantial context length of 32768 tokens, allowing for the paraphrasing of longer texts.
Good For
- Content Rephrasing: Generating alternative phrasings for sentences or paragraphs in various domains.
- Text Augmentation: Creating diverse versions of text for data augmentation in NLP tasks.
- Academic and Technical Writing: Assisting in rephrasing complex ideas or avoiding plagiarism by generating structurally different but semantically equivalent sentences.
- Improving Readability: Rewording text to enhance clarity or target different audiences.