Fine-Tune GPT-4o, Because One Size Doesn't Fit All
We can (finally) fine-tune custom versions of GPT-4o
At a Glance
OpenAI has launched fine-tuning capabilities for GPT-4o, allowing developers to customize the model to better suit specific applications. As one of the most requested features from practitioners, the AI giant is also offering organizations 1M training tokens per day for free through the end of September!
Deeper Learning
Why Should I Care About Fine-Tuning?: The fine-tuning feature for GPT-4o allows developers to adjust the model’s behavior and output according to specific needs. By training the model on a narrower set of data, organizations can improve its performance in specialized tasks, such as customer service, technical support, or industry-specific applications.
SOTA Performance: OpenAI tested fine-tuning on GPT-4o with select partners, including Cosine’s Genie, an AI software engineering assistant. Genie, powered by a fine-tuned GPT-4o model, autonomously identifies and resolves bugs, builds features, and refactors code with higher accuracy. The model was trained on real software engineering examples, enabling it to output specific formats like code patches. As a result, Genie achieved state-of-the-art (SOTA) scores of 43.8% on the SWE-bench Verified benchmark and 30.08% on SWE-bench Full, marking the largest improvement ever recorded on this benchmark.
Getting Started: OpenAI has made the fine-tuning process accessible to a wide range of users through the platform. You can visit the fine-tuning dashboard, create a new project, and select gpt-4o-2024-08-06 from the model selection menu. GPT-4o fine-tuning costs $25 per million tokens, with inference pricing set at $3.75 per million input tokens and $15 per million output tokens.
Privacy + Security: The AI startup claims that fine-tuned models remain entirely under the user's control, with full ownership of business data, including inputs and outputs. There are also safety evaluations ran on fine-tuned models to make sure they are not being misused.
So What?
Although this will require a massive dataset and is somewhat expensive, fine-tuning GPT-4o has the potential to yield very powerful but custom applications. OpenAI continues to empower businesses and organizations to create more effective and efficient AI solutions.
References
Blog Post: Fine-tuning now available for GPT-4o
Image from: PCMag
Share this post!
Fine-Tune GPT-4o, Because One Size Doesn't Fit All
We can (finally) fine-tune custom versions of GPT-4o
At a Glance
OpenAI has launched fine-tuning capabilities for GPT-4o, allowing developers to customize the model to better suit specific applications. As one of the most requested features from practitioners, the AI giant is also offering organizations 1M training tokens per day for free through the end of September!
Deeper Learning
Why Should I Care About Fine-Tuning?: The fine-tuning feature for GPT-4o allows developers to adjust the model’s behavior and output according to specific needs. By training the model on a narrower set of data, organizations can improve its performance in specialized tasks, such as customer service, technical support, or industry-specific applications.
SOTA Performance: OpenAI tested fine-tuning on GPT-4o with select partners, including Cosine’s Genie, an AI software engineering assistant. Genie, powered by a fine-tuned GPT-4o model, autonomously identifies and resolves bugs, builds features, and refactors code with higher accuracy. The model was trained on real software engineering examples, enabling it to output specific formats like code patches. As a result, Genie achieved state-of-the-art (SOTA) scores of 43.8% on the SWE-bench Verified benchmark and 30.08% on SWE-bench Full, marking the largest improvement ever recorded on this benchmark.
Getting Started: OpenAI has made the fine-tuning process accessible to a wide range of users through the platform. You can visit the fine-tuning dashboard, create a new project, and select gpt-4o-2024-08-06 from the model selection menu. GPT-4o fine-tuning costs $25 per million tokens, with inference pricing set at $3.75 per million input tokens and $15 per million output tokens.
Privacy + Security: The AI startup claims that fine-tuned models remain entirely under the user's control, with full ownership of business data, including inputs and outputs. There are also safety evaluations ran on fine-tuned models to make sure they are not being misused.
So What?
Although this will require a massive dataset and is somewhat expensive, fine-tuning GPT-4o has the potential to yield very powerful but custom applications. OpenAI continues to empower businesses and organizations to create more effective and efficient AI solutions.
References
Blog Post: Fine-tuning now available for GPT-4o
Image from: PCMag
Share this post!
Fine-Tune GPT-4o, Because One Size Doesn't Fit All
We can (finally) fine-tune custom versions of GPT-4o
At a Glance
OpenAI has launched fine-tuning capabilities for GPT-4o, allowing developers to customize the model to better suit specific applications. As one of the most requested features from practitioners, the AI giant is also offering organizations 1M training tokens per day for free through the end of September!
Deeper Learning
Why Should I Care About Fine-Tuning?: The fine-tuning feature for GPT-4o allows developers to adjust the model’s behavior and output according to specific needs. By training the model on a narrower set of data, organizations can improve its performance in specialized tasks, such as customer service, technical support, or industry-specific applications.
SOTA Performance: OpenAI tested fine-tuning on GPT-4o with select partners, including Cosine’s Genie, an AI software engineering assistant. Genie, powered by a fine-tuned GPT-4o model, autonomously identifies and resolves bugs, builds features, and refactors code with higher accuracy. The model was trained on real software engineering examples, enabling it to output specific formats like code patches. As a result, Genie achieved state-of-the-art (SOTA) scores of 43.8% on the SWE-bench Verified benchmark and 30.08% on SWE-bench Full, marking the largest improvement ever recorded on this benchmark.
Getting Started: OpenAI has made the fine-tuning process accessible to a wide range of users through the platform. You can visit the fine-tuning dashboard, create a new project, and select gpt-4o-2024-08-06 from the model selection menu. GPT-4o fine-tuning costs $25 per million tokens, with inference pricing set at $3.75 per million input tokens and $15 per million output tokens.
Privacy + Security: The AI startup claims that fine-tuned models remain entirely under the user's control, with full ownership of business data, including inputs and outputs. There are also safety evaluations ran on fine-tuned models to make sure they are not being misused.
So What?
Although this will require a massive dataset and is somewhat expensive, fine-tuning GPT-4o has the potential to yield very powerful but custom applications. OpenAI continues to empower businesses and organizations to create more effective and efficient AI solutions.
References
Blog Post: Fine-tuning now available for GPT-4o
Image from: PCMag
Share this post!
