This is a guest blog post by the XLScout team.
XLScout, a Toronto-based leader based in AI in intellectual property (IP), has developed a powerful, unique embedded model called Paraembed 2.0, which stems from an ambitious collaboration with Haggingface’s expert support program. The collaboration focuses on applying cutting-edge AI technology and open source models to enhance understanding and analysis of complex patent documents, including patent-specific terms, contexts and relationships. This allows XLScout products to provide the best performance for drafting patent applications, patent invalidation searches, and ensuring ideas compared to previously available patents and literature.
By fine-tuning high-quality multi-domain patent data curated by human experts, the 2.0 Paramved 2.0 increases accuracy by 23% compared to its predecessor, 1.0 Paramed 1.0, released in October 2023.
A journey to enhanced patent analysis
Initially, XLScout looked at its own AI models for patent analysis, but found that these closed source models, such as GPT-4 and Text-debedding-Ada-002, struggle to capture the subtle context required for technical and professional claims.
By integrating open source models such as BGE-Base-V1.5, Llama 2 70B, Falcon 40B, MixTral 8x7B and fine-tuning their own patent data using guidance from Face, XLScout has achieved more tailored and performance solutions. This shift has allowed us to better understand complex technical concepts and terms, revolutionizing the analysis and understanding of technical documents and patents.
Work with embracing your face through our expert support program
Collaboration with a hugging face helped to improve the quality and performance of XLScout’s solution. Here is a detailed overview of how this partnership evolved and its impact.
Early Development and Testing: XLScout was first built and tested on a custom Torchserve Inference Server (GCP) on Google Cloud Platform (GCP) with distributed data parallel (DDP) to provide multiple replicas. By integrating ONNX optimizations, they achieved an embedding performance rate of around 300 per second. Enhanced model performance with fine tuning: Embedded models were performed using data curated by patent experts. Not only did this workflow allow for more accurate and context-related embedding, it also significantly improved performance metrics and ensured greater accuracy in detecting related prior art. High-throughput serving: By leveraging Face’s inference endpoint with load balancing load, XLScout offers an embedded model with text-embedded inference (TEI) for high-throughput use cases that run successfully in production. The solution now delivers impressive performance and offers around 2700 embeddings per second! LLM Prompts and Inference: Collaboration included efforts in LLM prompt engineering and inference, which improved the model’s ability to generate accurate, context-specific patent drafts. Rapid engineering was adopted for patent-constrained use cases, ensuring that prompts were consistent and resulted in comprehensive, legal patent documents. Fine tuning LLM with instruction data: Formatting and fine tuning of instruction data was implemented using Meta and Mistral models. This fine-tuning allows for more accurate and detailed generation of some of the patent launch processes, further improving the quality of the generated output.
The partnership with the embracing face is a game changer for XLScout, significantly improving the processing speed, accuracy and overall quality of the LLM-driven solution. This collaboration ensures that universities, law firms and other clients will benefit from cutting-edge AI technology, the efficiency and innovation of patent landscapes.
XLScout’s AI-based IP Solution
XLScout offers cutting-edge AI-driven solutions that dramatically improve the efficiency and accuracy of patent-related processes. Their solutions are widely used by companies, universities and law firms, streamlining different aspects of IP workflows, from novelty searching and invalidation research to patent drafting.
Novelty Checker LLM: Use cutting-edge LLM and generation AI to quickly navigate patents and non-patent literature and validate ideas. It provides a comprehensive list of ranked previous art references along with key feature analysis reports. This tool allows inventors, researchers and patent experts to ensure that the invention is novel by comparing it to an extensive corpus of existing literature and patents. Invalidator LLM: Uses advanced LLM and generation AI to perform patent invalidation searches at extraordinary speed and accuracy. Provides a detailed list of ranked previous art references and a key feature analysis report. This service is important for law firms and businesses to efficiently challenge and evaluate the effectiveness of their patents. Drafting LLM: An automatic patented application draft platform that leverages the power of LLMS and generation AI. Within minutes, generate accurate and high quality preliminary patent drafts, including comprehensive claims, abstracts, drawings, backgrounds and descriptions. This solution helps patent practitioners significantly reduce the time and effort required to create detailed and accurate patent applications.
Companies and universities benefit from ensuring that new research results are properly protected, promote innovation and filed high-quality patents. Law firms utilize XLScout’s solutions to provide superior service to their clients, improving the quality of their patent prosecution and litigation efforts.
Partnership for Innovation
Sandeep Agarwal, CEO of XLScout, said: “This partnership combines Face’s open source models, tools and unparalleled capabilities of embracing teams, combining deep patent expertise. By tweaking unique data to these models, we are poised to revolutionize the patent drafting, analysis and licensing revolution.”
The joint efforts of XLScout and Hugging Face include training an open source model for XLScout’s extensive patent data collection. This synergy leverages XLScout’s expert knowledge and advanced AI capabilities to hold your face, bringing uniquely optimized models for patent research. Users benefit from more informed decision-making and valuable insights derived from complex patent documents.
Commitment to innovation and future planning
As a pioneer in the application of AI to intellectual property, XLScout is dedicated to exploring new frontiers with AI-driven innovation. This collaboration is a key step in bridging the gap between cutting-edge AI and real-world applications in IP analytics.
XLScout and Hugging Face set new standards for patent analysis, driving innovation and shaping the future of intellectual property. I’m excited to continue this amazing journey together!
For more information about hugging your company’s Face expert support program, contact us here – our team can contact us to discuss your requirements!
