Skip to content

Commit bb47165

Browse files
DVampireDVampire
committed
update readme
1 parent 454ce28 commit bb47165

2 files changed

Lines changed: 98 additions & 82 deletions

File tree

‎README.md‎

Lines changed: 48 additions & 41 deletions
Original file line numberDiff line numberDiff line change
@@ -122,41 +122,6 @@ git clone https://huggingface.co/datasets/gaia-benchmark/GAIA
122122
python examples/run_gaia.py
123123
```
124124

125-
### Deploying Qwen Models via vLLM
126-
127-
#### Step 1: Launch the vLLM Inference Service
128-
129-
```bash
130-
nohup bash -c 'CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
131-
--model /input0/Qwen3-32B \
132-
--served-model-name Qwen \
133-
--host 0.0.0.0 \
134-
--port 8000 \
135-
--max-num-seqs 16 \
136-
--enable-auto-tool-choice \
137-
--tool-call-parser hermes \
138-
--tensor_parallel_size 2' > vllm_qwen.log 2>&1 &
139-
```
140-
141-
Update `.env`:
142-
143-
```bash
144-
QWEN_API_BASE=http://localhost:8000/v1
145-
QWEN_API_KEY="abc"
146-
```
147-
148-
#### Step 2: Launch the Agent Service
149-
150-
```bash
151-
python main.py
152-
```
153-
154-
Example command:
155-
156-
```bash
157-
Use deep_researcher_agent to search the latest papers on the topic of 'AI Agent' and then summarize it.
158-
```
159-
160125
## Experiments
161126

162127
We evaluated our agent on the GAIA validation set and achieved state-of-the-art performance on May 10th.
@@ -195,15 +160,57 @@ playwright install chromium --with-deps --no-shell
195160

196161
Function-calling is now supported natively by GPT-4.1 / Gemini 2.5 Pro. Claude-3.7-Sonnet is also recommended.
197162

163+
### 4. Use vllm for local models
164+
We provide huggingface as a shortcut to the local model. Also provide vllm as a way to start services so that parallel acceleration can be provided.
165+
166+
#### Step 1: Launch the vLLM Inference Service
167+
168+
```bash
169+
nohup bash -c 'CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
170+
--model /input0/Qwen3-32B \
171+
--served-model-name Qwen \
172+
--host 0.0.0.0 \
173+
--port 8000 \
174+
--max-num-seqs 16 \
175+
--enable-auto-tool-choice \
176+
--tool-call-parser hermes \
177+
--tensor_parallel_size 2' > vllm_qwen.log 2>&1 &
178+
```
179+
180+
Update `.env`:
181+
182+
```bash
183+
QWEN_API_BASE=http://localhost:8000/v1
184+
QWEN_API_KEY="abc"
185+
```
186+
187+
#### Step 2: Launch the Agent Service
188+
189+
```bash
190+
python main.py
191+
```
192+
193+
Example command:
194+
195+
```bash
196+
Use deep_researcher_agent to search the latest papers on the topic of 'AI Agent' and then summarize it.
197+
```
198+
198199
## Acknowledgement
199200

200-
DeepResearchAgent is inspired by and improved upon:
201+
DeepResearchAgent is primarily inspired by the architecture of smolagents. The following improvements have been made:
202+
- The codebase of smolagents has been modularized for better structure and organization.
203+
- The original synchronous framework has been refactored into an asynchronous one.
204+
- The multi-agent setup process has been optimized to make it more user-friendly and efficient.
205+
206+
We would like to express our gratitude to the following open source projects, which have greatly contributed to the development of this work:
207+
- [smolagents](https://github.com/huggingface/smolagents) - A lightweight agent framework.
208+
- [OpenManus](https://github.com/mannaandpoem/OpenManus) - An asynchronous agent framework.
209+
- [browser-use](https://github.com/browser-use/browser-use) - An AI-powered browser automation tool.
210+
- [crawl4ai](https://github.com/unclecode/crawl4ai) - A web crawling library for AI applications.
211+
- [markitdown](https://github.com/microsoft/markitdown) - A tool for converting files to Markdown format.
201212

202-
* [smolagents](https://github.com/huggingface/smolagents)
203-
* [OpenManus](https://github.com/mannaandpoem/OpenManus)
204-
* [browser-use](https://github.com/browser-use/browser-use)
205-
* [crawl4ai](https://github.com/unclecode/crawl4ai)
206-
* [markitdown](https://github.com/microsoft/markitdown)
213+
We sincerely appreciate the efforts of all contributors and maintainers of these projects for their commitment to advancing AI technologies and making them available to the wider community.
207214

208215
## Contribution
209216

‎README_CN.md‎

Lines changed: 50 additions & 41 deletions
Original file line numberDiff line numberDiff line change
@@ -117,41 +117,6 @@ git clone https://huggingface.co/datasets/gaia-benchmark/GAIA
117117
python examples/run_gaia.py
118118
```
119119

120-
### 示例 3：部署本地 Qwen 模型（使用 vLLM）
121-
122-
#### 第一步：启动 vLLM 服务
123-
124-
```bash
125-
nohup bash -c 'CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
126-
--model /input0/Qwen3-32B \
127-
--served-model-name Qwen \
128-
--host 0.0.0.0 \
129-
--port 8000 \
130-
--max-num-seqs 16 \
131-
--enable-auto-tool-choice \
132-
--tool-call-parser hermes \
133-
--tensor_parallel_size 2' > vllm_qwen.log 2>&1 &
134-
```
135-
136-
`.env` 文件配置：
137-
138-
```bash
139-
QWEN_API_BASE=http://localhost:8000/v1
140-
QWEN_API_KEY="abc"
141-
```
142-
143-
#### 第二步：启动智能体服务
144-
145-
```bash
146-
python main.py
147-
```
148-
149-
输入任务命令示例：
150-
151-
```bash
152-
Use deep_researcher_agent to search the latest papers on the topic of 'AI Agent' and then summarize it.
153-
```
154-
155120
## 实验结果
156121

157122
我们在 GAIA 验证集上进行了评估，并于 5 月 10 日达到最先进性能：
@@ -188,15 +153,59 @@ playwright install chromium --with-deps --no-shell
188153

189154
推荐使用 Claude-3.7-Sonnet 作为规划智能体，现已兼容 GPT-4.1 和 Gemini-2.5-Pro。
190155

156+
### 4. 使用 vllm 进行本地模型部署
157+
我们提供 huggingface 作为加载本地模型的快捷方式。我们还提供 vllm 作为启动服务的方式，以便提供并行加速。
158+
159+
### 部署本地 Qwen 模型（使用 vLLM）
160+
161+
#### 第一步：启动 vLLM 服务
162+
163+
```bash
164+
nohup bash -c 'CUDA_VISIBLE_DEVICES=0,1 python -m vllm.entrypoints.openai.api_server \
165+
--model /input0/Qwen3-32B \
166+
--served-model-name Qwen \
167+
--host 0.0.0.0 \
168+
--port 8000 \
169+
--max-num-seqs 16 \
170+
--enable-auto-tool-choice \
171+
--tool-call-parser hermes \
172+
--tensor_parallel_size 2' > vllm_qwen.log 2>&1 &
173+
```
174+
175+
`.env` 文件配置：
176+
177+
```bash
178+
QWEN_API_BASE=http://localhost:8000/v1
179+
QWEN_API_KEY="abc"
180+
```
181+
182+
#### 第二步：启动智能体服务
183+
184+
```bash
185+
python main.py
186+
```
187+
188+
输入任务命令示例：
189+
190+
```bash
191+
Use deep_researcher_agent to search the latest papers on the topic of 'AI Agent' and then summarize it.
192+
```
193+
191194
## 致谢
192195

193-
本项目受以下开源项目启发，并在此基础上做出改进：
196+
DeepResearchAgent 主要借鉴了 smolagents 的架构设计，并在此基础上做出了以下改进：
197+
- 对 smolagents 的代码进行了模块化重构，使其结构更加清晰和易于维护。
198+
- 将原有的同步执行框架改为异步执行框架。
199+
- 优化了多智能体的初始化与部署流程，使其更加高效且易用。
200+
201+
我们由衷感谢以下开源项目对本项目的重要贡献：
202+
- [smolagents](https://github.com/huggingface/smolagents) - 轻量级智能体框架。
203+
- [OpenManus](https://github.com/mannaandpoem/OpenManus) - 异步智能体框架。
204+
- [browser-use](https://github.com/browser-use/browser-use) - 基于 AI 的浏览器自动化工具。
205+
- [crawl4ai](https://github.com/unclecode/crawl4ai) - 面向 AI 应用的网页爬取库。
206+
- [markitdown](https://github.com/microsoft/markitdown) - 文件转 Markdown 工具。
194207

195-
* [smolagents](https://github.com/huggingface/smolagents)
196-
* [OpenManus](https://github.com/mannaandpoem/OpenManus)
197-
* [browser-use](https://github.com/browser-use/browser-use)
198-
* [crawl4ai](https://github.com/unclecode/crawl4ai)
199-
* [markitdown](https://github.com/microsoft/markitdown)
208+
我们由衷感谢上述项目的所有贡献者和维护者，感谢他们推动 AI 技术进步并将其开放给社区使用。
200209

201210
## 贡献
202211

0 commit comments

Comments
 (0)