sglang

SGLang

面向大模型和 VLM 的高性能服务框架，借助 RadixAttention 实现自动前缀缓存。

何时使用 SGLang

在以下情况下使用 SGLang：

• 需要结构化输出（JSON、正则、语法）
• 构建带有重复前缀（系统提示词、工具）的 Agent
• 带函数调用的 Agentic 工作流
• 共享上下文的多轮对话
• 需要更快的 JSON 解码（相比标准方式快 3×）

在以下情况下改用 vLLM：

• 无需结构的简单文本生成
• 不需要前缀缓存
• 想要成熟、经过广泛测试的生产系统

在以下情况下改用 TensorRT-LLM：

• 追求单请求延迟最低（无需批处理）
• 仅限 NVIDIA 部署
• 需要在 H100 上进行 FP8/INT4 量化

快速开始

安装

# pip install (recommended)
pip install "sglang[all]"

# With FlashInfer (faster, CUDA 11.8/12.1)
pip install sglang[all] flashinfer -i https://flashinfer.ai/whl/cu121/torch2.4/

# From source
git clone https://github.com/sgl-project/sglang.git
cd sglang
pip install -e "python[all]"

启动服务

# Basic server (Llama 3-8B)
python -m sglang.launch_server \
    --model-path meta-llama/Meta-Llama-3-8B-Instruct \
    --port 30000

# With RadixAttention (automatic prefix caching)
python -m sglang.launch_server \
    --model-path meta-llama/Meta-Llama-3-8B-Instruct \
    --port 30000 \
    --enable-radix-cache  # Default: enabled

# Multi-GPU (tensor parallelism)
python -m sglang.launch_server \
    --model-path meta-llama/Meta-Llama-3-70B-Instruct \
    --tp 4 \
    --port 30000

基础推理

import sglang as sgl

# Set backend
sgl.set_default_backend(sgl.OpenAI("http://localhost:30000/v1"))

# Simple generation
@sgl.function
def simple_gen(s, question):
    s += "Q: " + question + "\n"
    s += "A:" + sgl.gen("answer", max_tokens=100)

# Run
state = simple_gen.run(question="What is the capital of France?")
print(state["answer"])
# Output: "The capital of France is Paris."

结构化 JSON 输出

import sglang as sgl

@sgl.function
def extract_person(s, text):
    s += f"Extract person information from: {text}\n"
    s += "Output JSON:\n"

    # Constrained JSON generation
    s += sgl.gen(
        "json_output",
        max_tokens=200,
        regex=r'\{"name": "[^"]+", "age": \d+, "occupation": "[^"]+"\}'
    )

# Run
state = extract_person.run(
    text="John Smith is a 35-year-old software engineer."
)
print(state["json_output"])
# Output: {"name": "John Smith", "age": 35, "occupation": "software engineer"}

RadixAttention（核心创新）

它做什么：自动缓存并复用跨请求的公共前缀。

性能：

• 对于共享系统提示词的 Agentic 工作负载，快 5×
• 对于重复示例的少样本提示，快 10×
• 零配置 - 自动生效

工作原理：

1. 为所有已处理的 token 构建基数树
2. 自动检测共享前缀
3. 为匹配的前缀复用 KV 缓存
4. 只计算新的 token

示例（带系统提示词的 Agent）：

Request 1: [SYSTEM_PROMPT] + "What's the weather?"
→ Computes full prompt (1000 tokens)

Request 2: [SAME_SYSTEM_PROMPT] + "Book a flight"
→ Reuses system prompt KV cache (998 tokens)
→ Only computes 2 new tokens
→ 5× faster!

结构化生成模式

带 schema 的 JSON

@sgl.function
def structured_extraction(s, article):
    s += f"Article: {article}\n\n"
    s += "Extract key information as JSON:\n"

    # JSON schema constraint
    schema = {
        "type": "object",
        "properties": {
            "title": {"type": "string"},
            "author": {"type": "string"},
            "summary": {"type": "string"},
            "sentiment": {"type": "string", "enum": ["positive", "negative", "neutral"]}
        },
        "required": ["title", "author", "summary", "sentiment"]
    }

    s += sgl.gen("info", max_tokens=300, json_schema=schema)

state = structured_extraction.run(article="...")
print(state["info"])
# Output: Valid JSON matching schema

正则约束生成

@sgl.function
def extract_email(s, text):
    s += f"Extract email from: {text}\n"
    s += "Email: "

    # Email regex pattern
    s += sgl.gen(
        "email",
        max_tokens=50,
        regex=r'[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}'
    )

state = extract_email.run(text="Contact [email protected] for details")
print(state["email"])
# Output: "[email protected]"

基于语法的生成

@sgl.function
def generate_code(s, description):
    s += f"Generate Python code for: {description}\n"
    s += "```python\n"

    # EBNF grammar for Python
    python_grammar = """
    ?start: function_def
    function_def: "def" NAME "(" [parameters] "):" suite
    parameters: parameter ("," parameter)*
    parameter: NAME
    suite: simple_stmt | NEWLINE INDENT stmt+ DEDENT
    """

    s += sgl.gen("code", max_tokens=200, grammar=python_grammar)
    s += "\n```"

带函数调用的 Agent 工作流

import sglang as sgl

# Define tools
tools = [
    {
        "name": "get_weather",
        "description": "Get weather for a location",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {"type": "string"}
            }
        }
    },
    {
        "name": "book_flight",
        "description": "Book a flight",
        "parameters": {
            "type": "object",
            "properties": {
                "from": {"type": "string"},
                "to": {"type": "string"},
                "date": {"type": "string"}
            }
        }
    }
]

@sgl.function
def agent_workflow(s, user_query, tools):
    # System prompt (cached with RadixAttention)
    s += "You are a helpful assistant with access to tools.\n"
    s += f"Available tools: {tools}\n\n"

    # User query
    s += f"User: {user_query}\n"
    s += "Assistant: "

    # Generate with function calling
    s += sgl.gen(
        "response",
        max_tokens=200,
        tools=tools,  # SGLang handles tool call format
        stop=["User:", "\n\n"]
    )

# Multiple queries reuse system prompt
state1 = agent_workflow.run(
    user_query="What's the weather in NYC?",
    tools=tools
)
# First call: Computes full system prompt

state2 = agent_workflow.run(
    user_query="Book a flight to LA",
    tools=tools
)
# Second call: Reuses system prompt (5× faster)

性能基准

RadixAttention 加速

少样本提示（提示中包含 10 个示例）：

• vLLM：2.5 秒/请求
• SGLang：0.25 秒/请求（快 10×）
• 吞吐量：高 4×

Agent 工作流（1000 token 系统提示词）：

• vLLM：1.8 秒/请求
• SGLang：0.35 秒/请求（快 5×）

JSON 解码：

• 标准：45 tok/s
• SGLang：135 tok/s（快 3×）

吞吐量（Llama 3-8B，A100）

工作负载	vLLM	SGLang	加速
简单生成	2500 tok/s	2800 tok/s	1.12×
少样本（10 个示例）	500 tok/s	5000 tok/s	10×
Agent（工具调用）	800 tok/s	4000 tok/s	5×
JSON 输出	600 tok/s	2400 tok/s	4×

多轮对话

@sgl.function
def multi_turn_chat(s, history, new_message):
    # System prompt (always cached)
    s += "You are a helpful AI assistant.\n\n"

    # Conversation history (cached as it grows)
    for msg in history:
        s += f"{msg['role']}: {msg['content']}\n"

    # New user message (only new part)
    s += f"User: {new_message}\n"
    s += "Assistant: "
    s += sgl.gen("response", max_tokens=200)

# Turn 1
history = []
state = multi_turn_chat.run(history=history, new_message="Hi there!")
history.append({"role": "User", "content": "Hi there!"})
history.append({"role": "Assistant", "content": state["response"]})

# Turn 2 (reuses Turn 1 KV cache)
state = multi_turn_chat.run(history=history, new_message="What's 2+2?")
# Only computes new message (much faster!)

# Turn 3 (reuses Turn 1 + Turn 2 KV cache)
state = multi_turn_chat.run(history=history, new_message="Tell me a joke")
# Progressively faster as history grows

高级功能

投机解码

# Launch with draft model (2-3× faster)
python -m sglang.launch_server \
    --model-path meta-llama/Meta-Llama-3-70B-Instruct \
    --speculative-model meta-llama/Meta-Llama-3-8B-Instruct \
    --speculative-num-steps 5

多模态（视觉模型）

@sgl.function
def describe_image(s, image_path):
    s += sgl.image(image_path)
    s += "Describe this image in detail: "
    s += sgl.gen("description", max_tokens=200)

state = describe_image.run(image_path="photo.jpg")
print(state["description"])

批处理与并行请求

# Automatic batching (continuous batching)
states = sgl.run_batch(
    [
        simple_gen.bind(question="What is AI?"),
        simple_gen.bind(question="What is ML?"),
        simple_gen.bind(question="What is DL?"),
    ]
)

# All 3 processed in single batch (efficient)

OpenAI 兼容 API

# Start server with OpenAI API
python -m sglang.launch_server \
    --model-path meta-llama/Meta-Llama-3-8B-Instruct \
    --port 30000

# Use with OpenAI client
curl http://localhost:30000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "default",
    "messages": [
      {"role": "system", "content": "You are helpful"},
      {"role": "user", "content": "Hello"}
    ],
    "temperature": 0.7,
    "max_tokens": 100
  }'

# Works with OpenAI Python SDK
from openai import OpenAI
client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY")

response = client.chat.completions.create(
    model="default",
    messages=[{"role": "user", "content": "Hello"}]
)

支持的模型

文本模型：

• Llama 2、Llama 3、Llama 3.1、Llama 3.2
• Mistral、Mixtral
• Qwen、Qwen2、QwQ
• DeepSeek-V2、DeepSeek-V3
• Gemma、Phi-3

视觉模型：

• LLaVA、LLaVA-OneVision
• Phi-3-Vision
• Qwen2-VL

来自 HuggingFace 的 100+ 模型

硬件支持

NVIDIA：A100、H100、L4、T4（CUDA 11.8+） AMD：MI300、MI250（ROCm 6.0+） Intel：带 GPU 的 Xeon（即将推出） Apple：通过 MPS 支持 M1/M2/M3（实验性）

参考资料

• 结构化生成指南 - JSON schema、正则、语法、验证
• RadixAttention 深入解析 - 工作原理、优化、基准
• 生产部署 - 多 GPU、监控、自动扩缩容

资源

• GitHub：https://github.com/sgl-project/sglang
• 文档：https://sgl-project.github.io/
• 论文：RadixAttention（arXiv:2312.07104）
• Discord：https://discord.gg/sglang