Microsoft has introduced a new way to fine-tune its Phi-3 small language model without the hassle of managing servers, and it’s free initially. Phi-3 is a 3 billion parameter model designed for third-party developers to create new applications at a low cost. It offers affordable performance on coding, common sense reasoning, and general knowledge. The model comes in 6 variations with different numbers of parameters and context lengths, ranging from $0.0003 to $0.0005 per 1,000 input tokens.
Previously, developers had to set up their own servers or run the model on local machines for fine-tuning. However, Microsoft has now made serverless fine-tuning available through its Azure AI development platform. This allows developers to quickly start AI development without managing infrastructure. Phi-3 models can be fine-tuned for specific use cases, such as learning new skills or improving response quality.
The pricing for serverless fine-tuning of Phi-3-mini-4k-instruct starts at $0.004 per 1,000 tokens. This move by Microsoft not only benefits developers in its ecosystem but also competes with OpenAI’s efforts to attract enterprise AI developers. OpenAI recently announced free fine-tuning of GPT-4o mini for certain users, while Meta and Mistral have also released models that can be fine-tuned for different uses.
Overall, the AI landscape is evolving rapidly, with providers offering various models and options for enterprise development. Developers now have more choices and opportunities to create innovative AI applications tailored to their specific needs.