FinetuneDB

FinetuneDB is an AI fine tuning platform for building custom language models from a team's own data. It combines dataset creation, model evaluation, production logging and prompt management in one workspace, with the aim of better performance at lower cost than generic models. It is built for engineers, product managers and domain experts. A collaborative editor lets teams build proprietary fine tuning datasets together. Evaluators and metrics measure output quality, speed and token usage, using human and AI feedback, and a Copilot feature automates evaluations and model improvements. Production logging captures user interactions and model responses, with filters and tracing of nested actions across language chains. A prompt playground supports creating and managing prompts collaboratively. Serverless inference and fine tuning run on open models including Llama and Mixtral families. The Basic plan costs $0 per month for one user, with one project and dataset, 1000 logs per month and $10 of free credit. Pro is $50 per month from two users, with unlimited projects, datasets and logs, custom evaluation workflows, integrations such as vLLM, 24 hour support response and $100 of free credit. Custom plans add model hosting, unlimited users, role based access, SAML single sign on and SOC 2 compliance. Serverless inference is pay as you go per million tokens, for example $0.30 for input on Llama 3 8B. The site announces that FinetuneDB has been acquired by Opper AI.