Haven

Haven, located in Berlin, Germany, operates within the B2B sector, specifically focusing on engineering, product, and design. The company, which is part of the Y Combinator S23 batch, offers a platform that enables companies to fine-tune and deploy open-source LLMs effortlessly with a single API call. Customers can personalize these models based on their data and choose to run them securely on-premise or in the cloud. Haven’s platform is open-source and self-hostable, ensuring that users maintain ownership and control over their LLMs. The company provides a free start option to make its technology accessible, supports its users with documentation and a blog, and is fully licensed under Apache-2.0. Pricing for fine-tuning runs is based on the number of tokens in the training dataset, with specific rates for models like HuggingFaceH4/zephyr-7b-beta and meta-llama/Llama-2-7b-chat-hf set at $0.004 per 1K tokens. This cost increases if training spans multiple epochs.