OpenAI introduces fine-tuning capabilities for GPT-3.5 Turbo

A fine-tuned version of GPT-3.5 Turbo can outperform GPT-4, said OpenAI.
Sejal Sharma
Sam Altman

US-based AI company OpenAI just released the fine-tuning API for GPT-3.5 Turbo. This gives developers more flexibility to customize models that perform better for their use cases. The company ran tests, showing that the fine-tuned versions of GPT-3.5 Turbo can surpass GPT-4's base capabilities on certain tasks.

And coming this fall for GPT-4. OpenAI said in the announcement that developers will retain full ownership of the data, not OpenAI or any other company.

OpenAI released gpt-3.5-turbo in March this year as a ChatGPT model family for various non-chat uses. It’s priced at $0.002 per 1k tokens, which the AI company claims is 10 times cheaper than GPT-3.5 models.

Empowering devs to customize models for their needs

Not to be mistaken with an update, the technique of fine-tuning in large language models (LLMs) and machine learning is used to customize a pre-trained model to perform specific tasks and run it at scale. Developers will be able to leverage the data that the model has already been trained on to apply it to a particular field or task of their choice with better results.

"Fine-tuning is most powerful when combined with other techniques such as prompt engineering, information retrieval, and function calling," said the company.

For example, a company requires a chatbot service that can reply in English and Spanish. So, they would also see a model already trained on answering prompts in English to train/fine-tune it to answer prompts in Spanish.

Other uses involve copywriting for advertising, tailored customer service, generating code, and focused text summarization.

Delivering a developer's ask

OpenAI said that ever since the release of GPT-3.5 Turbo, developers and businesses have asked for the ability to customize the model to create unique and differentiated experiences for their users.

Moreover, the company has claimed that fine-tuning GPT-3.5 Turbo has allowed businesses to reduce their prompt size by up to 90 percent while maintaining performance, leading to faster API calls and reduced costs.

One X user said that using a fine-tuned GPT 3.5 Turbo is 8 times the cost of generating from the base model - GPT 3.5. The user said that the developer has "to be in the “reduce prompt size by 90 percent” bucket they mention to get cost-effectiveness out of it."

“Fine-tuning with GPT-3.5-Turbo can also handle 4k tokens—double our previous fine-tuned models,” said OpenAI. GPT-3.5-Turbo prices, as published by OpenAI, are $0.0080 per 1K tokens for training, $0.0120 per 1K tokens for input usage, and $0.0120 per 1K tokens for output usage.

Other models with OpenAI available for fine-tuning are babbage-002 and davinci-002, the pricing for which can be found here.

For more on fine-tuning GPT-3.5-Turbo, see OpenAI’s help guides here.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board