OpenAI also clarified that data used in the fine-tuning API is not used to train the model. The company will charge developers for initial training and usage, with costs detailed per 1,000 tokens. OpenAI has also partnered with Scale to assist companies in customizing models using the new fine-tuning feature.
Key takeaways:
- OpenAI has launched fine-tuning for GPT-3.5 Turbo and plans to launch for GPT-4 this fall, allowing developers to customize models for their specific use cases.
- Fine-tuning involves teaching the model a task by showing it many examples, and is recommended when there are a lot of examples to be used in the prompts and when the model is run often and needs to run fast.
- OpenAI will charge developers two types of costs for using fine-tuning: an initial training cost and a usage cost, with specific rates outlined for different types of usage.
- Following the launch of fine-tuning for GPT-3.5, OpenAI has announced Scale as its preferred partner to help companies use it to customize models.