Together AI has joined forces with Adaption, a startup focused on AI data optimization, to integrate its fine-tuning platform directly into Adaption’s offering. This collaboration aims to accelerate the creation of customized open-source AI models.
Adaption, co-founded by former Cohere and Google DeepMind leaders Sara Hooker and Sudip Roy, provides tools for analyzing dataset structure, adapting examples, and evaluating data quality, reporting an average 82% increase in data quality for early users.
With this new integration, users of Adaption’s platform can connect their Together AI accounts to seamlessly transition from data optimization to model fine-tuning. The workflow allows users to refine their training datasets within Adaption and then initiate fine-tuning jobs on Together AI’s infrastructure, leveraging optimized hyperparameters.
Sara Hooker, Co-founder & CEO of Adaption, stated, "Together Fine-Tuning gives Adaptive Data users the infrastructure to turn shaped datasets into stronger, more reliable open models." The platform supports various fine-tuning methods like LoRA and full fine-tuning, along with large models exceeding 100 billion parameters.
Together AI’s fine-tuning service offers features such as cost estimation before training, real-time progress tracking, and direct model export to Hugging Face Hub, aiming to simplify the process for developers and researchers.