Content
summary Summary

Microsoft and OpenAI now provide fine-tuning services for their recently introduced GPT-4o mini model and Phi-3. This allows developers to customize these models for specific applications.

Ad

OpenAI states that fine-tuning for GPT-4o mini costs 50 percent less than for GPT-3.5 Turbo. As part of the launch, OpenAI is offering free fine-tuning with up to two million training tokens daily until September 23, 2024.

Compared to GPT-3.5 Turbo, GPT-4o mini is also 90 percent cheaper for input tokens and 80 percent cheaper for output tokens. The training context is four times longer than GPT-3.5 Turbo at 65,000 tokens, while the inference context is eight times longer at 128,000 tokens.

The new GPT-4o mini model also offers more performance and supports image processing, although fine-tuning is currently limited to text. Fine-tuning is available to OpenAI level 4 and 5 developers (minimum $250 paid and 14+ days since first successful payment) through the Fine-Tuning dashboard.

Ad
Ad

At the same time, Microsoft has introduced serverless fine-tuning services for the Phi 3-mini and Phi 3-medium models in Azure AI. According to Microsoft, this allows developers to quickly tune models for cloud and edge scenarios without requiring extensive computing resources.

Microsoft also reports "significant improvements in core quality, instruction-following, and structured output" through updates for Phi-3-mini.

LLM market sees increasing price pressure

OpenAI's discount campaign also highlights the changing language model market: More models from more providers, some open source, are competing for a limited number of application scenarios in a market that generates significant interest but little revenue.

Competition is currently intensifying for both small and large models, which have seen only incremental improvements since the introduction of the GPT-4 class. The result is a price war.

Ad
Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Summary
  • Microsoft and OpenAI are now offering fine-tuning services for their recently introduced GPT-4o mini and Phi-3 models, allowing developers to customize these models for specific applications at a lower cost compared to GPT-3.5 Turbo.
  • GPT-4o mini also offers more performance, image processing support, and a longer training and inference context than GPT-3.5 Turbo. Fine-tuning for visual tasks is not yet available. Microsoft says that Phi-3-mini got an update that increased performance.
  • OpenAI is offering free fine-tuning with up to two million training tokens daily until September 23, 2024, as part of the launch.
Sources
Online journalist Matthias is the co-founder and publisher of THE DECODER. He believes that artificial intelligence will fundamentally change the relationship between humans and computers.
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.