Established by Mira Murati, the former CTO of OpenAI, the AI firm Thinking Machines Lab has officially rolled out its inaugural product, Tinker, to the general public. Tinker represents an API meticulously crafted for the fine-tuning of language models. It offers support for fine-tuning a diverse array of models, spanning from compact to expansive Mixture-of-Experts (MoE) models (such as Qwen-235B-A22B). Users have the flexibility to effortlessly transition between different model scales by merely adjusting strings within the Python code. Moreover, Tinker furnishes low-level operational primitives along with an open-source support library. Its compatibility with the OpenAI API allows for seamless calls and validation throughout the model training journey.
