OpenAI has announced that third-party developers can now fine-tune its newly launched multimodal large model, GPT-4o. This feature allows developers to adjust the model's behavior according to their application needs, such as altering the tone, following specific instructions, or improving accuracy in technical tasks. Even with minimal training data, significant performance improvements can be achieved through fine-tuning.
For developers, using this new feature is straightforward. They simply need to access OpenAI's fine-tuning dashboard, click "Create," and then select "gpt-4o-2024-08-06" from the base model dropdown menu. This ease of operation means that fine-tuning can greatly enhance the model's performance in various fields, whether it be programming or creative writing.
Access: https://platform.openai.com/finetune
Free 1 Million Tokens Daily Until September 23
OpenAI is also offering a special promotion: until September 23, 2024, developers will receive up to 1 million free Tokens daily. These Tokens are the basic units for the model to understand and process information, and developers need to convert their business-related data into Tokens for effective fine-tuning.
It's important to note that the usual cost for fine-tuning the GPT-4o model is $25 per 1 million Tokens, while the cost for using the fine-tuned model for inference is $3.75 per 1 million input Tokens and $15 per 1 million output Tokens.
For the smaller GPT-4o mini model, OpenAI provides 2 million free Tokens daily until the same deadline. This measure not only attracts a large number of developers but also strengthens OpenAI's market competitiveness against competitors like Google, Anthropic, and newly launched open-source models such as Nous Research's Hermes3.
Additionally, while promoting the fine-tuning feature, OpenAI has not overlooked issues of security and data privacy. They ensure that the fine-tuned model can fully control enterprise data, guaranteeing that inputs and outputs are not used to train other models. To ensure compliance with OpenAI's usage policies, the company has implemented multi-layered security measures.
Successful Fine-Tuning Cases
There have been some successful fine-tuning cases, such as an AI software engineering company named Cosine, which achieved a top score of 43.8% on the SWE-bench benchmark through fine-tuning.
Another AI solutions company named Distyl achieved an execution accuracy of 71.83% on the BIRD-SQL benchmark by fine-tuning GPT-4o. These successful examples undoubtedly add more credibility to OpenAI's fine-tuning feature.
OpenAI is striving to realize a vision where every organization has its own customized AI model, and this fine-tuning feature is a significant step towards that goal.
Key Points:
🌟 Fine-Tuning Feature Launch: OpenAI introduces the fine-tuning feature for the GPT-4o model, allowing developers to adjust the model's behavior according to their needs.
💰 Free Tokens Giveaway: Developers can receive 1 million free Tokens daily for model fine-tuning, attracting many developers to participate.
🔒 Data Security Assurance: OpenAI emphasizes data privacy and security, ensuring that fine-tuned models do not use input and output data for retraining.