OpenPipe, a Seattle, WA-based tool that lets developers create their own application-specific Large Language Models (LLMs), closed a $6.7m seed funding round.
The round was led by Costanoa Ventures, with participation from Y Combinator and prominent angel investors, including Logan Kilpatrick, former head of developer relations at OpenAI, Alex Graveley, creator of GitHub Copilot, and Tom Preston-Werner, co-founder of GitHub.
The company intends to use the funds to scale its team and accelerate product development.
OpenPipe’s key insight is that smaller models are able to match or exceed the performance of much larger ones when specialized to a specific task. This specialization is done through a process known as “fine-tuning.” Fine-tuning massively improves response quality and correctness, but many developers aren’t able to perform this task because of the complex workflow required. Existing tools are designed for engineers with specialized knowledge and skills in machine learning, which limits teams and organizations without a dedicated specialist. OpenPipe provides an easy-to-use tool that enables everyday developers to train and deploy fine-tuned models faster, more cost-effectively and with greater accuracy.
OpenPipe works by integrating into a customer’s existing codebase to collect their existing prompts, which are then used as training data to fine-tune a much smaller, more specialized model. Developers can then use the fine-tuned model as a drop-in replacement for their existing prompts.
Co-founded by Kyle Corbitt and David Corbitt, OpenPipe is a tool that simplifies Large Language Model (LLM) integration for developers in production applications. Its platform allows developers to fine-tune models for accuracy, reducing latency and costs, enabling full-stack and AI engineers and developers to train and deploy fine-tuned models without relying on data science and machine learning specialists. Its active customer base includes startups and enterprises that use fine-tuned models to perform business workflows, like translating and extracting information from government regulations, categorizing and filtering YouTube videos based on transcripts or routing inbound customer service requests to the appropriate teams.
The company is partnered with several industry leaders to provide fine-tuning like Langfuse and Athina AI for data collection, and OctoAI and Fireworks AI to serve the completed models.
FinSMEs
26/03/2024