Shulou(Shulou.com)11/24 Report--

CTOnews.com August 23, OpenAI announced on its website that it will introduce GPT-3.5Turbo fine-tuning and update API so that enterprises and developers can use their own data to build their own ChatGPT combined with business use cases.

OpenAI claims that such a customized model can meet or exceed the capabilities of GPT-4 on certain tasks. The company also said that GPT-4 's fine-tuning function will be launched this fall.

OpenAI wrote in a blog post: "since the launch of GPT-3.5 Turbo, developers and enterprises have asked to be able to customize models to create a unique and differentiated experience for users. This update enables developers to customize models that better suit their use cases and run them on a large scale."

With fine tuning, companies that use GPT-3.5 Turbo through OpenAI's API can make the model better follow instructions, such as having it always respond in a given language. Alternatively, they can improve the model's ability to consistently format responses (for example, to complete code snippets) and hone the "feel" of the model output, such as its tone, so that it is better suited to the brand.

In addition, fine-tuning enables OpenAI customers to shorten text prompts to speed up API calls and reduce costs. "early testers reduced the size of prompts by 90 per cent by fine-tuning the instructions of the model itself," OpenAI claimed in a blog post.

Currently, fine-tuning requires preparing data, uploading necessary files, and creating fine-tuning jobs through OpenAI's API. The company said all fine-tuning data must go through the audit system supported by "audit" API and GPT-4 to see if it violates OpenAI's security standards. However, OpenAI plans to introduce a fine-tuning UI in the future, which includes a dashboard to check the status of ongoing fine-tuning workloads.

The fine-tuning costs are as follows:

Training: $0.008 / 1K tokens

Use input: $0.012 / 1K tokens

Use output: $0.016 / 1K tokens

OpenAI said that the GPT-3.5-turbo fine-tuning of training files containing 100000 token (about 75000 words) would cost about $2.40 (CTOnews.com Note: currently about 18 yuan).

In addition, OpenAI today provides two updated basic GPT-3 models (babbage-002 and davinci-002) that can also be fine-tuned to support paging and "more extensibility". As previously announced, OpenAI plans to deactivate the original GPT-3 base model on January 4, 2024.

OpenAI said that fine-tuning support for GPT-4 (unlike GPT-3.5, where GPT-4 can understand images other than text) will be released later this fall, but did not provide any other details.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.