OpenAI introduces an enhanced initiative, the Custom Model program, catering to enterprise clients seeking tailor-made generative AI solutions for their specific needs.
Following its successful launch at the inaugural DevDay conference last year, the program has seen considerable uptake, prompting OpenAI to expand its offerings for optimized performance.
A notable addition is the Assisted Fine-Tuning component, which goes beyond conventional fine-tuning methods. Employing advanced techniques and additional hyperparameters, this feature allows organizations to establish robust data training pipelines and evaluation systems, thereby enhancing model performance for targeted tasks.
Furthermore, the introduction of Custom-Trained Models provides clients with bespoke AI solutions built upon OpenAI’s base models and tools like GPT-4. This enables deep fine-tuning and integration of domain-specific knowledge, as exemplified by SK Telecom’s enhancement of GPT-4 for telecom-related conversations in Korean and Harvey’s development of a custom model for case law analysis.
New fine-tuning API features
Today, the team introduced new features to give developers even more control over their fine-tuning jobs, including:
- Epoch-based Checkpoint Creation: Automatically produce one full fine-tuned model checkpoint during each training epoch, which reduces the need for subsequent retraining, especially in the cases of overfitting
- Comparative Playground: A new side-by-side Playground UI for comparing model quality and performance, allowing human evaluation of the outputs of multiple models or fine-tune snapshots against a single prompt
- Third-party Integration: Support for integrations with third-party platforms (starting with Weights and Biases this week) to let developers share detailed fine-tuning data to the rest of their stack
- Comprehensive Validation Metrics: The ability to compute metrics like loss and accuracy over the entire validation dataset instead of a sampled batch, providing better insight on model quality
- Hyperparameter Configuration: The ability to configure available hyperparameters from the Dashboard (rather than only through the API or SDK)
- Fine-Tuning Dashboard Improvements: Including the ability to configure hyperparameters, view more detailed training metrics, and rerun jobs from previous configurations
OpenAI envisions a future where personalized AI models tailored to industry-specific needs become the norm for organizations of all sizes. Such customization not only fosters more impactful AI implementations but also addresses the growing demand for generative AI solutions.
With OpenAI’s revenue reportedly nearing $2 billion annually, there’s an internal drive to sustain momentum, especially as the company collaborates with Microsoft on a proposed $100 billion data center. Offering consulting services like custom model training serves to fuel revenue growth while addressing the ongoing challenges of training and serving flagship AI models.
New Custom Models Program
- Assisted Fine-Tuning
At DevDay last November, OpenAI announced a Custom Model program designed to train and optimize models for a specific domain, in partnership with a dedicated group of OpenAI researchers. Since then, we’ve met with dozens of customers to assess their custom model needs and evolved our program to further maximize performance.
Today, they introduced their assisted fine-tuning offering as part of the Custom Model program. Assisted fine-tuning is a collaborative effort with our technical teams to leverage techniques beyond the fine-tuning API, such as additional hyperparameters and various parameter efficient fine-tuning (PEFT) methods at a larger scale.
- Custom-Trained Model
In some cases, organizations need to train a purpose-built model from scratch that understands their business, industry, or domain. Fully custom-trained models imbue new knowledge from a specific domain by modifying key steps of the model training process using novel mid-training and post-training techniques.
Organizations that see success with a fully custom-trained model often have large quantities of proprietary data—millions of examples or billions of tokens—that they want to use to teach the model new knowledge or complex, unique behaviors for highly specific use cases.
Additionally, tailored models alleviate strain on OpenAI’s infrastructure by being more efficient and performant compared to generic counterparts. As demand for generative AI continues to surge, customized solutions emerge as an attractive proposition, aligning with OpenAI’s commitment to innovation and scalability.
In tandem with the expanded Custom Model program, OpenAI introduces new features for fine-tuning GPT-3.5, including a user-friendly dashboard for model comparison and performance analysis, seamless integration with third-party platforms like Weights & Biases, and enhanced tooling capabilities.
While details on fine-tuning for GPT-4 remain undisclosed since its early access launch at DevDay, OpenAI’s ongoing efforts signal a commitment to advancing AI technology and meeting the evolving needs of its clientele.