Tensorfuse
Tensorfuse, based in San Francisco, CA, is a B2B company specializing in testing and evaluation platforms for LLM applications, offering services like offline and online evaluations to improve LLM application retrieval and deployment.
Company Overview
Tensorfuse, formerly known as 10K and TensorFuse, is located in San Francisco, CA, USA. The company operates in the B2B industry, specifically focusing on Engineering, Product, and Design. With a small team size of 2, Tensorfuse is part of the Y Combinator W24 batch. The company offers services across the United States, America/Canada, and remotely. Their mission is to help developers identify failure points in their LLM pipelines and provide ways to fix them.
Services
Tensorfuse offers a comprehensive testing and evaluation platform for LLM applications. This includes both offline and online evaluations. Offline evaluations involve experimenting with different prompts, architectures, and models on production datasets. Online evaluations, on the other hand, include live production monitoring to prevent security threats and performance degradation. The company provides an end-to-end testing platform for offline evaluations, supporting dataset creation, automated testing, and continuous deployment. Their in-house preference models for online evaluations are 100x cost-effective and 10x faster than GPT-3.5/4.
Subscription Plans
Tensorfuse offers various subscription plans, catering to different needs. The 'Hobby' plan is designed for individual developers working on side projects. This plan includes evaluating up to 1,000 logs per month, data retention for up to 1 month, and basic support. For larger teams with specific needs, Tensorfuse provides an 'Enterprise' plan. This plan includes enterprise-wide deployment, on-prem deployment of evaluator models, SSO + SAML, and a dedicated customer support representative. All plans start with a 2-week free trial.
Blog and Resources
Tensorfuse maintains a blog that features articles related to improving LLM application retrieval and production deployment. These resources are valuable for developers looking to enhance the performance and reliability of their LLM applications. The blog offers insights, tips, and best practices to help users optimize their LLM pipelines and ensure their applications perform as expected in production environments.
Technology and Innovation
Tensorfuse aims to bring predictability to every company's LLM app development cycle. The company provides both offline and online evaluations that help developers spot and rectify failure points in their LLM pipelines. Their preference models used for online evaluations offer a cost-effective and faster alternative to popular models such as GPT-3.5/4. Tensorfuse's services are specifically geared towards creating a predictable and efficient development process for LLM applications.