www.analyticsdrift.com
Image source: Analytics Drift
OpenAI has updated its AI capabilities by introducing fine-tuning features for third-party developers in its most advanced large multimodal model (LMM), GPT-4o.
Image source: OpenAI
Fine-tuning is the process by which developers can customize pre-trained AI models according to their requirements to optimize the model to perform specific tasks.
Image source: Canva
According to OpenAI, developers can now fine-tune GPT-4o with custom datasets to create a high-performing AI model at low costs. This will benefit small businesses with limited data.
Image source: Canva
Developers can visit OpenAI’s fine-tuning dashboard, click Create, and choose GPT-4o-2024-08-06 from the base model drop-down list to access the fine-tuning feature.
Image source: Openai
Developers can access the fine-tuning feature in all the paid versions of GPT-4o. For promotion, OpenAI will also offer free 1M training tokens to every organization up to September 23.
Image source: Canva
Tokens are the numerical formats of text representing concepts learned by LLM. For fine-tuning, a developer needs to convert relevant data into tokens that the LLM can understand.
Image source: Canva
Cosine achieved 43.8% on the SWE bench benchmark for fine-tuning GPT-4o for its AI agent Genie. Distyl’s fine-tuned GPT-4o ranked first on the BIRD-SQL benchmark with 71.83% accuracy.
Image source: Openai
OpenAI assures businesses of full control over their dataset, which will not be used to train other models, and also offers automated evaluations and usage monitoring for safety.
Image source: Canva