limecoding/gemma2-2b-it-finetuned-patent

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2.6BQuant:BF16Context Size:8kPublished:Oct 1, 2024License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The limecoding/gemma2-2b-it-finetuned-patent model is a 2.6 billion parameter Gemma2-based instruction-tuned language model developed by limecoding. It is specifically fine-tuned to assist with drafting patent specifications from a general invention description. This model leverages LoRA for efficient training and utilizes a dataset combining research paper summaries and patent claims from KIPRIS. Its primary strength lies in generating structured and semantically similar patent-related text.

Loading preview...

Model Overview

The limecoding/gemma2-2b-it-finetuned-patent is a 2.6 billion parameter instruction-tuned model, developed by limecoding, based on the Gemma2 architecture. It is specifically designed and fine-tuned to generate patent specifications from a given invention description. The model was efficiently trained using Unsloth and Hugging Face's TRL library, leveraging LoRA (Low-Rank Adaptation) for fine-tuning.

Key Capabilities

  • Patent Specification Drafting: Excels at generating structured and semantically relevant text for patent specifications, including invention titles, technical fields, and claims, based on a general description.
  • Specialized Knowledge: Trained on a unique dataset comprising research paper summaries from AI-Hub and patent claims data from KIPRIS (Korea Intellectual Property Rights Information Service), providing it with specialized domain understanding.
  • Efficient Training: Utilizes Unsloth for faster training and reduced VRAM consumption, allowing for larger batch sizes.

Use Cases

This model is particularly well-suited for:

  • Automated Patent Drafting: Assisting legal professionals and inventors in quickly generating initial drafts of patent specifications.
  • Intellectual Property Research: Extracting and structuring key information from invention descriptions into patent-like formats.
  • Patent Document Generation: Creating components of patent documents, such as claims and technical field descriptions, from raw input.