openai-api - OpenAI ChatGPT (GPT-3.5) API : Can I fine-tune the gpt-3. 5-turbo 模型？

Question

我有一个包含大量数据的 SQL 表，需要使用 Chat Completion API 将 SQL 表数据训练到 ChatGPT。

我尝试使用 ChatGPT 生成 SQL 查询，但这无法按预期工作。有时它会生成不适当的查询。

Answer 1

更新:2023 年 8 月 22 日

现在可以对 GPT-3.5 进行微调，如官方 OpenAI blog 中所述:

Fine-tuning for GPT-3.5 Turbo is now available, with fine-tuning forGPT-4 coming this fall. This update gives developers the ability tocustomize models that perform better for their use cases and run thesecustom models at scale. Early tests have shown a fine-tuned version ofGPT-3.5 Turbo can match, or even outperform, base GPT-4-levelcapabilities on certain narrow tasks. As with all our APIs, data sentin and out of the fine-tuning API is owned by the customer and is notused by OpenAI, or any other organization, to train other models.

此外，还引入了两个新模型(即 davinci-002 和 babbage-002)作为 GPT-3 模型 (davinci) 的替代品>、curie、babbage 和 ada)，因为这些 GPT-3 模型将于 2024 年 1 月 4 日关闭。因此，如果您看看官方OpenAI documentation现在，您将看到以下内容:

What models can be fine-tuned?

Fine-tuning is currently available for the following models:

• gpt-3.5-turbo-0613 (recommended)
• babbage-002
• davinci-002

We expectgpt-3.5-turbo to be the right model for most users in terms of resultsand ease of use, unless you are migrating a legacy fine-tuned model.

注意:GPT-3 模型已从列表中删除，但在 2024 年 1 月 4 日之前应该仍可进行微调。

<小时/>

您无法微调 gpt-3.5-turbo 模型。

正如官方所说OpenAI documentation :

What models can be fine-tuned?

Fine-tuning is currently only availablefor the following base models: davinci, curie, babbage, and ada. Theseare the original models that do not have any instruction followingtraining (like text-davinci-003 does for example). You are also ableto continue fine-tuning a fine-tuned model to add additional datawithout having to start from scratch.