first draft

Signed-off-by: guangli.bao <guangli.bao@daocloud.io>

Author: tukwila

docs/assets/sample-output1.png

@@ -0,0 +1,117 @@
+# GuideLLM Benchmark Testing Best Practice
+
+Do first easy-go guidellm benchmark testing from scratch using vLLM Simulator.
+
+## 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 Startup OpenAI-compatible API in vLLM simulator docker container
+
+```bash
+docker pull ghcr.io/llm-d/llm-d-inference-sim:v0.4.0
+
+docker run --rm --publish 8000:8000 \
+ghcr.io/llm-d/llm-d-inference-sim:v0.4.0  \
+--port 8000 \
+--model "Qwen/Qwen2.5-1.5B-Instruct"  \
+--lora-modules '{"name":"tweet-summary-0"}' '{"name":"tweet-summary-1"}'
+```
+
+For more detailed instructions, refer to: [vLLM Simulator](https://llm-d.ai/docs/architecture/Components/inference-sim)
+
+Docker image versions: [Docker Images](https://github.com/llm-d/llm-d-inference-sim/pkgs/container/llm-d-inference-sim)
+
+Check open-ai api working via curl:
+
+- check /v1/models
+
+```bash
+curl --request GET 'http://localhost:8000/v1/models'
+```
+
+- check /v1/chat/completions
+
+```bash
+curl --request POST 'http://localhost:8000/v1/chat/completions' \
+--header 'Content-Type: application/json' \
+--data-raw '{
+    "model": "tweet-summary-0",
+    "stream": false,
+    "messages": [{"role": "user", "content": "Say this is a test!"}]
+}'
+```
+
+- check /v1/completions
+
+```bash
+curl --request POST 'http://localhost:8000/v1/completions' \
+--header 'Content-Type: application/json' \
+--data-raw '{
+    "model": "tweet-summary-0",
+    "stream": false,
+    "prompt": "Say this is a test!",
+    "max_tokens": 128
+}'
+```
+
+#### 1.5 Download Tokenizer
+
+Download Qwen/Qwen2.5-1.5B-Instruct tokenizer files from [Qwen/Qwen2.5-1.5B-Instruct](https://modelscope.cn/models/Qwen/Qwen2.5-1.5B-Instruct/files) save to local path such as ${local_path}/Qwen2.5-1.5B-Instruct
+
+```bash
+ls ./Qwen2.5-1.5B-Instruct
+merges.txt              tokenizer.json          tokenizer_config.json   vocab.json
+```
+
+______________________________________________________________________
+
+## 🚀 2. Running Benchmarks
+
+```bash
+guidellm benchmark \
+--target "http://localhost:8000/" \
+--model "tweet-summary-0" \
+--processor "${local_path}/Qwen2.5-1.5B-Instruct" \
+--rate-type sweep \
+--max-seconds 10 \
+--max-requests 10 \
+--data "prompt_tokens=128,output_tokens=56"
+```
+
+______________________________________________________________________
+
+## 📊 3. Results Interpretation
+
+![alt text](../assets/sample-output1.png) ![alt text](../assets/sample-output2.png) ![alt text](../assets/sample-output3.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 complete.