Haven
Haven is a Berlin-based B2B company specializing in engineering, product, and design, providing tools to fine-tune, evaluate, and deploy open-source large language models (LLMs) with flexible deployment options.
Services
Haven enables companies to fine-tune and deploy open source large language models (LLMs) with a single API call. Their platform allows customers to personalize models on their data and securely run them on-premise or in the cloud. They offer tools for training, evaluating, and deploying LLMs that users own. Additionally, they provide free start options for new users and are fully Apache-2.0 licensed.
Industry and Sub-Industry
Haven operates in the B2B sector, specifically within the engineering, product, and design sub-industries. They cater to companies across Germany and Europe, providing specialized tools to manage and deploy open source LLMs efficiently.
Y Combinator S23 Batch
Haven is part of the Y Combinator Summer 2023 (S23) batch. This association with one of the world's most prestigious startup accelerators highlights their potential for innovation and growth within the B2B engineering sector.
Open-Source Platform
Haven's platform is open-source and self-hostable, allowing businesses to have full control over their deployed models. They provide comprehensive documentation and a blog for user support, ensuring that customers are well-equipped to utilize their tools effectively.
Pricing and Licensing
Haven's pricing for fine-tuning runs depends on the number of tokens in the training dataset. The cost for models like HuggingFaceH4/zephyr-7b-beta and meta-llama/Llama-2-7b-chat-hf is $0.004 per 1,000 tokens. If training occurs across multiple epochs, the token price is multiplied by the number of epochs. The platform is fully licensed under Apache-2.0, offering flexibility and transparency for users.