|
|
---
|
|
|
title: OpenAI GPT 系列 Tools Calling 评测
|
|
|
description: >-
|
|
|
使用 LobeChat 测试 OpenAI GPT 系列模型（GPT 3.5-turbo / GPT-4 /GPT-4o） 的工具调用（Function
|
|
|
Calling）能力，并展现评测结果
|
|
|
tags:
|
|
|
- Tools Calling
|
|
|
- Benchmark
|
|
|
- Function Calling
|
|
|
- 工具调用
|
|
|
- 插件
|
|
|
---
|
|
|
|
|
|
# OpenAI GPT Series Tool Calling
|
|
|
|
|
|
Overview of the Tool Calling capabilities of OpenAI GPT series models:
|
|
|
|
|
|
| Model | Tool Calling Support | Streaming | Parallel | Simple Instruction Score | Complex Instruction Score |
|
|
|
| ------------- | -------------------- | --------- | -------- | ------------------------ | ------------------------- |
|
|
|
| GPT-3.5-turbo | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟 |
|
|
|
| GPT-4-turbo | ✅ | ✅ | ✅ | 🌟🌟 | 🌟🌟 |
|
|
|
| GPT-4o | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
|
|
|
|
|
|
<Callout type={'info'}>
|
|
|
For testing instructions, see [Tools Calling - Evaluation Task
|
|
|
Introduction](/docs/usage/tools-calling#evaluation-task-introduction)
|
|
|
</Callout>
|
|
|
|
|
|
## GPT 3.5-turbo
|
|
|
|
|
|
### Simple Instruction Call: Weather Inquiry
|
|
|
|
|
|
Test Instruction: Instruction ①
|
|
|
|
|
|
<Video src="https://github.com/lobehub/lobe-chat/assets/28616219/65901ee2-78b8-4f56-9e0d-6407c484f434" />
|
|
|
|
|
|
<Image alt="Tool Calling for Simple Instruction in GPT 3.5 Turbo" src="https://github.com/lobehub/lobe-chat/assets/28616219/1251dfc0-d1c4-4c3d-825e-dd6205793d53" />
|
|
|
|
|
|
<details>
|
|
|
<summary>Streaming Tool Calling Raw Output:</summary>
|
|
|
</details>
|
|
|
|
|
|
### Complex Instruction Call: Wenshengtu
|
|
|
|
|
|
Test Instruction: Instruction ②
|
|
|
|
|
|
<Video src="https://github.com/lobehub/lobe-chat/assets/28616219/2047665f-ab22-4da7-a390-0fb4ec5a2a14" />
|
|
|
|
|
|
<Image alt="Tool Calling for Complex Instruction in GPT 3.5 Turbo" src="https://github.com/lobehub/lobe-chat/assets/28616219/125ad028-a621-4433-b5fa-321f8fd76302" />
|
|
|
|
|
|
<details>
|
|
|
<summary>Streaming Tool Calling Raw Output:</summary>
|
|
|
</details>
|
|
|
|
|
|
## GPT-4 Turbo
|
|
|
|
|
|
### Simple Instruction Call: Weather Inquiry
|
|
|
|
|
|
Test Instruction: Instruction ①
|
|
|
|
|
|
Unlike GPT-3.5 Turbo, GPT-4 Turbo did not respond with "okay" when calling Tool Calling, and after multiple tests, it remained the same. Therefore, in this follow-up of a compound instruction, it is not as good as GPT-3.5 Turbo, but the remaining two capabilities are still good.
|
|
|
|
|
|
Of course, it is also possible that GPT-4 Turbo's model has more "autonomy" and believes that it does not need to output this "okay."
|
|
|
|
|
|
<Video src="https://github.com/lobehub/lobe-chat/assets/28616219/f865d91b-b84a-4258-ae09-9d1e15eeb43d" />
|
|
|
|
|
|
<Image alt="Tool Calling for Simple Instruction in GPT-4 Turbo" src="https://github.com/lobehub/lobe-chat/assets/28616219/19298693-7a9b-4b54-9e28-c46b541b4f41" />
|
|
|
|
|
|
<details>
|
|
|
<summary>Streaming Tool Calling Raw Output:</summary>
|
|
|
</details>
|
|
|
|
|
|
### Complex Instruction Call: Wenshengtu
|
|
|
|
|
|
Test Instruction: Instruction ②
|
|
|
|
|
|
<Video src="https://github.com/lobehub/lobe-chat/assets/28616219/69989faf-9b98-41ec-ba51-40cc3545d8d1" />
|
|
|
|
|
|
<Image alt="Tool Calling for Complex Instruction in GPT-4 Turbo" src="https://github.com/lobehub/lobe-chat/assets/28616219/8329c1b2-5e36-4457-946c-ce3781b05afd" />
|
|
|
|
|
|
<details>
|
|
|
<summary>Streaming Tool Calling Raw Output:</summary>
|
|
|
</details>
|
|
|
|
|
|
## GPT-4o
|
|
|
|
|
|
### Simple Instruction Call: Weather Inquiry
|
|
|
|
|
|
Test Instruction: Instruction ①
|
|
|
|
|
|
Similar to GPT-3.5, GPT-4o performs very well in following compound instructions in simple instruction calls.
|
|
|
|
|
|
<Video src="https://github.com/lobehub/lobe-chat/assets/28616219/c77b65ab-0854-4e1f-a25b-ff43275bd318" />
|
|
|
|
|
|
<Image alt="Tool Calling for Simple Instruction in GPT-4o" src="https://github.com/lobehub/lobe-chat/assets/28616219/e5d6214f-f628-4064-a330-cbd7c5d474ac" />
|
|
|
|
|
|
<details>
|
|
|
<summary>Streaming Tool Calling Raw Output:</summary>
|
|
|
</details>
|
|
|
|
|
|
### Complex Instruction Call: Wenshengtu
|
|
|
|
|
|
Test Instruction: Instruction ②
|
|
|
|
|
|
<Video src="https://github.com/lobehub/lobe-chat/assets/28616219/714bd86a-3b58-4941-8323-186c3fa4c6ea" />
|
|
|
|
|
|
<Image alt="Tool Calling for Complex Instruction in GPT-4o" src="https://github.com/lobehub/lobe-chat/assets/28616219/8329c1b2-5e36-4457-946c-ce3781b05afd" />
|
|
|
|
|
|
<details>
|
|
|
<summary>Streaming Tool Calling Raw Output:</summary>
|
|
|
|
|
|
```yml
|
|
|
```
|
|
|
</details>
|
