first practice example

Signed-off-by: guangli.bao <[email protected]>

Author: tukwila

docs/examples/benchmark_practice_on_drun.md

@@ -0,0 +1,85 @@
+# GuideLLM Benchmark Testing Best Practices
+
+[https://console.d.run/](https://console.d.run/) is one AI infrastructure platform in which there are some deployed model playgrounds. Now, one `guidellm` benchmark testing practice example will be based on one chat model.
+
+## Getting Started
+
+### 📦 1. Benchmark Testing Environment Setup
+
+#### 1.1 Create a Conda Environment (recommended)
+
+```bash
+conda create -n guidellm-bench python=3.11 -y
+conda activate guidellm-bench
+```
+
+#### 1.2 Install Dependencies
+
+```bash
+git clone https://github.com/vllm-project/guidellm.git
+cd guidellm
+pip install guidellm
+```
+
+For more detailed instructions, refer to [GuideLLM README](https://github.com/vllm-project/guidellm/blob/main/README.md).
+
+#### 1.3 Verify Installation
+
+```bash
+guidellm --help
+```
+
+#### 1.4 Apply for Account and API Key in D.run
+
+Firstly, register an account, refer to [D.run Registration](https://docs.d.run/en/#register-account); then, create an API key, refer to [D.run API Key](https://docs.d.run/en/#register-account); finally, charge your account at [D.run Account Management](https://docs.d.run/en/#register-account).
+
+#### 1.5 Chat with Model in D.run
+
+Check if you can use the chat model in D.run.
+
+![alt text](image.png)
+
+#### 1.6 Find Out the HTTP Request URL and Body
+
+Use the Developer Tool in Chrome browser or press F12 to open Network, then chat with the LLM model to capture the HTTP request URL and body.
+
+![alt text](image-1.png)
+
+![alt text](image-2.png)
+
+In this request, the vllm backend service URL is `https://chat.d.run`; vllm model is `public/qwen2.5-72b-instruct-awq`. These two pieces of information will be used in the following benchmark command.
+
+#### 1.7 Download a Chat Dataset from Modelscope
+
+Download the chat dataset JSON file `Open-Source-Meeseeks-high-quality.json` from [Modelscope - Meeseeks](https://modelscope.cn/datasets/meituan/Meeseeks/files).
+
+![alt text](image-3.png)
+
+---
+
+## 🚀 2. Running Benchmarks
+
+```bash
+export GUIDELLM__OPENAI__API_KEY="${api_key}"
+guidellm benchmark \
+ --target "https://chat.d.run/" \
+ --model "public/qwen2.5-72b-awq" \
+ --rate-type "throughput" \
+ --data-args '{"prompt_column": "prompt", "split": "train"}' \
+ --max-requests 10 \
+ --data "/{$local_path}/Open-Source-Meeseeks-high-quality.json"
+```
+
+---
+
+## 📊 3. Results Interpretation
+
+![alt text](image-4.png)
+
+After the benchmark completes, key results are clear and straightforward, such as:
+
+* **`TTFT`**: Time to First Token
+* **`TPOT`**: Time Per Output Token
+* **`ITL`**: Inter-Token Latency
+
+The first benchmark test is complete.