README.md 24.5 KB
Newer Older
1
<div align="center"  id="sglangtop">
Kushal Agrawal's avatar
Kushal Agrawal committed
2
<img src="https://raw.githubusercontent.com/sgl-project/sglang/main/assets/logo.png" alt="logo" width="400" margin="10px"></img>
Lianmin Zheng's avatar
Lianmin Zheng committed
3

Yineng Zhang's avatar
Yineng Zhang committed
4
5
6
7
8
9
[![PyPI](https://img.shields.io/pypi/v/sglang)](https://pypi.org/project/sglang)
![PyPI - Downloads](https://img.shields.io/pypi/dm/sglang)
[![license](https://img.shields.io/github/license/sgl-project/sglang.svg)](https://github.com/sgl-project/sglang/tree/main/LICENSE)
[![issue resolution](https://img.shields.io/github/issues-closed-raw/sgl-project/sglang)](https://github.com/sgl-project/sglang/issues)
[![open issues](https://img.shields.io/github/issues-raw/sgl-project/sglang)](https://github.com/sgl-project/sglang/issues)

Yineng Zhang's avatar
Yineng Zhang committed
10
11
</div>

Lianmin Zheng's avatar
Lianmin Zheng committed
12
13
--------------------------------------------------------------------------------

Lianmin Zheng's avatar
Lianmin Zheng committed
14
15
| [**Blog**](https://lmsys.org/blog/2024-07-25-sglang-llama3/) | [**Paper**](https://arxiv.org/abs/2312.07104) | [**Slides**](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_dev_day_v2.pptx) | [**Join Slack**](https://join.slack.com/t/sgl-fru7574/shared_invite/zt-2ngly9muu-t37XiH87qvD~6rVBTkTEHw) |
[**Join Bi-Weekly Development Meeting (Oct. 19)**](https://calendar.app.google/GYW7S8QGoanCuaxW6) |
Lianmin Zheng's avatar
Lianmin Zheng committed
16

Ying Sheng's avatar
Ying Sheng committed
17
## Upcoming Events
18
- [Oct. 16, 2024] Online meetup for efficient LLM deployment and serving, co-hosted by SGLang, FlashInfer, and MLC LLM! Fill out the [Google form](https://forms.gle/B3YeedLxmrrhL1NM8) to receive the invite link. (Directly join the [Meeting link](https://us06web.zoom.us/j/81267838086?pwd=qz1s2pktTsl9xNPg2lXM7fMR2A0mOm.1) if you are too late to register.)
19

Lianmin Zheng's avatar
Lianmin Zheng committed
20
## News
Yineng Zhang's avatar
Yineng Zhang committed
21
- [2024/09] 🔥 SGLang v0.3 Release: 7x Faster DeepSeek MLA, 1.5x Faster torch.compile, Multi-Image/Video LLaVA-OneVision ([blog](https://lmsys.org/blog/2024-09-04-sglang-v0-3/)).
Ying Sheng's avatar
Ying Sheng committed
22
23
- [2024/07] 🔥 Faster Llama3 Serving with SGLang Runtime (vs. TensorRT-LLM, vLLM) ([blog](https://lmsys.org/blog/2024-07-25-sglang-llama3/)).
- [2024/02] SGLang enables **3x faster JSON decoding** with compressed finite state machine ([blog](https://lmsys.org/blog/2024-02-05-compressed-fsm/)).
Ying Sheng's avatar
Ying Sheng committed
24

Ying Sheng's avatar
Ying Sheng committed
25
26
27
<details>
<summary>More</summary>

28
- [2024/04] SGLang is used by the official **LLaVA-NeXT (video)** release ([blog](https://llava-vl.github.io/blog/2024-04-30-llava-next-video/)).
Ying Sheng's avatar
Ying Sheng committed
29
- [2024/01] SGLang provides up to **5x faster inference** with RadixAttention ([blog](https://lmsys.org/blog/2024-01-17-sglang/)).
Ying Sheng's avatar
Ying Sheng committed
30
31
32
33
- [2024/01] SGLang powers the serving of the official **LLaVA v1.6** release demo ([usage](https://github.com/haotian-liu/LLaVA?tab=readme-ov-file#demo)).

</details>

Ying Sheng's avatar
Ying Sheng committed
34
35
36
37
38
39
40
## About
SGLang is a fast serving framework for large language models and vision language models.
It makes your interaction with models faster and more controllable by co-designing the backend runtime and frontend language.
The core features include:

- **Fast Backend Runtime**: Provides efficient serving with RadixAttention for prefix caching, jump-forward constrained decoding, continuous batching, token attention (paged attention), tensor parallelism, FlashInfer kernels, chunked prefill, and quantization (INT4/FP8/AWQ/GPTQ).
- **Flexible Frontend Language**: Offers an intuitive interface for programming LLM applications, including chained generation calls, advanced prompting, control flow, multi-modal inputs, parallelism, and external interactions.
Lianmin Zheng's avatar
Lianmin Zheng committed
41
- **Extensive Model Support**: Supports a wide range of generative models (Llama, Gemma, Mistral, QWen, DeepSeek, LLaVA, etc.) and embedding models (e5-mistral), with easy extensibility for integrating new models.
Ying Sheng's avatar
Ying Sheng committed
42
43
- **Active Community**: SGLang is open-source and backed by an active community with industry adoption.

Lianmin Zheng's avatar
Lianmin Zheng committed
44
45
46
## Contents
- [Install](#install)
- [Backend: SGLang Runtime (SRT)](#backend-sglang-runtime-srt)
Ying Sheng's avatar
Ying Sheng committed
47
- [Frontend: Structured Generation Language (SGLang)](#frontend-structured-generation-language-sglang)
Lianmin Zheng's avatar
Lianmin Zheng committed
48
49
50
51
52
53
- [Benchmark And Performance](#benchmark-and-performance)
- [Roadmap](#roadmap)
- [Citation And Acknowledgment](#citation-and-acknowledgment)

## Install

54
55
You can install SGLang using any of the methods below.

Lianmin Zheng's avatar
Lianmin Zheng committed
56
57
### Method 1: With pip
```
58
pip install --upgrade pip
Lianmin Zheng's avatar
Lianmin Zheng committed
59
pip install "sglang[all]"
Lianmin Zheng's avatar
Lianmin Zheng committed
60

Lianmin Zheng's avatar
Lianmin Zheng committed
61
# Install FlashInfer CUDA kernels
62
pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.4/
Lianmin Zheng's avatar
Lianmin Zheng committed
63
```
64

Lianmin Zheng's avatar
Lianmin Zheng committed
65
### Method 2: From source
Lianmin Zheng's avatar
Lianmin Zheng committed
66
```
Ying Sheng's avatar
Ying Sheng committed
67
# Use the last release branch
Lianmin Zheng's avatar
Lianmin Zheng committed
68
git clone -b v0.3.3.post1 https://github.com/sgl-project/sglang.git
Lianmin Zheng's avatar
Lianmin Zheng committed
69
70
cd sglang

Lianmin Zheng's avatar
Lianmin Zheng committed
71
pip install --upgrade pip
Lianmin Zheng's avatar
Lianmin Zheng committed
72
73
pip install -e "python[all]"

Lianmin Zheng's avatar
Lianmin Zheng committed
74
# Install FlashInfer CUDA kernels
75
pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.4/
Lianmin Zheng's avatar
Lianmin Zheng committed
76
```
77

Lianmin Zheng's avatar
Lianmin Zheng committed
78
### Method 3: Using docker
79
The docker images are available on Docker Hub as [lmsysorg/sglang](https://hub.docker.com/r/lmsysorg/sglang/tags), built from [Dockerfile](https://github.com/sgl-project/sglang/tree/main/docker).
80
Replace `<secret>` below with your huggingface hub [token](https://huggingface.co/docs/hub/en/security-tokens).
Ying Sheng's avatar
Ying Sheng committed
81

Liangsheng Yin's avatar
Liangsheng Yin committed
82
83
84
85
```bash
docker run --gpus all \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
86
    --env "HF_TOKEN=<secret>" \
Liangsheng Yin's avatar
Liangsheng Yin committed
87
88
    --ipc=host \
    lmsysorg/sglang:latest \
89
    python3 -m sglang.launch_server --model-path meta-llama/Llama-3.1-8B-Instruct --host 0.0.0.0 --port 30000
Liangsheng Yin's avatar
Liangsheng Yin committed
90
91
```

92
93
### Method 4: Using docker compose

94
<details>
Dr. Artificial曾小健's avatar
Dr. Artificial曾小健 committed
95
<summary>More</summary>
96

97
> This method is recommended if you plan to serve it as a service.
98
> A better approach is to use the [k8s-sglang-service.yaml](docker/k8s-sglang-service.yaml).
99

100
1. Copy the [compose.yml](docker/compose.yaml) to your local machine
101
2. Execute the command `docker compose up -d` in your terminal.
102
</details>
103

104
105
### Method 5: Run on Kubernetes or Clouds with SkyPilot

106
<details>
Dr. Artificial曾小健's avatar
Dr. Artificial曾小健 committed
107
<summary>More</summary>
108

109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
To deploy on Kubernetes or 12+ clouds, you can use [SkyPilot](https://github.com/skypilot-org/skypilot).

1. Install SkyPilot and set up Kubernetes cluster or cloud access: see [SkyPilot's documentation](https://skypilot.readthedocs.io/en/latest/getting-started/installation.html).
2. Deploy on your own infra with a single command and get the HTTP API endpoint:
<details>
<summary>SkyPilot YAML: <code>sglang.yaml</code></summary>

```yaml
# sglang.yaml
envs:
  HF_TOKEN: null

resources:
  image_id: docker:lmsysorg/sglang:latest
  accelerators: A100
  ports: 30000

run: |
  conda deactivate
  python3 -m sglang.launch_server \
129
    --model-path meta-llama/Llama-3.1-8B-Instruct \
130
131
132
    --host 0.0.0.0 \
    --port 30000
```
133
</details>
134
135
136
137
138
139
140
141
142

```bash
# Deploy on any cloud or Kubernetes cluster. Use --cloud <cloud> to select a specific cloud provider.
HF_TOKEN=<secret> sky launch -c sglang --env HF_TOKEN sglang.yaml

# Get the HTTP API endpoint
sky status --endpoint 30000 sglang
```
3. To further scale up your deployment with autoscaling and failure recovery, check out the [SkyServe + SGLang guide](https://github.com/skypilot-org/skypilot/tree/master/llm/sglang#serving-llama-2-with-sglang-for-more-traffic-using-skyserve).
143
</details>
144
145


Lianmin Zheng's avatar
Lianmin Zheng committed
146
### Common Notes
Lianmin Zheng's avatar
Lianmin Zheng committed
147
- [FlashInfer](https://github.com/flashinfer-ai/flashinfer) is the default attention kernel backend. It only supports sm75 and above. If you encounter any FlashInfer-related issues on sm75+ devices (e.g., T4, A10, A100, L4, L40S, H100), please switch to other kernels by adding `--attention-backend triton --sampling-backend pytorch` and open an issue on GitHub.
Lianmin Zheng's avatar
Lianmin Zheng committed
148
- If you only need to use the OpenAI backend, you can avoid installing other dependencies by using `pip install "sglang[openai]"`.
Ying Sheng's avatar
Ying Sheng committed
149

Ying Sheng's avatar
Ying Sheng committed
150
151
152
## Backend: SGLang Runtime (SRT)
The SGLang Runtime (SRT) is an efficient serving engine.

Ying Sheng's avatar
Ying Sheng committed
153
### Quick Start
Ying Sheng's avatar
Ying Sheng committed
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
Launch a server
```
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000
```

Send a request
```
curl http://localhost:30000/generate \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Once upon a time,",
    "sampling_params": {
      "max_new_tokens": 16,
      "temperature": 0
    }
  }'
```
Lianmin Zheng's avatar
Lianmin Zheng committed
171
172

Learn more about the argument specification, streaming, and multi-modal support [here](docs/en/sampling_params.md).
Ying Sheng's avatar
Ying Sheng committed
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201

### OpenAI Compatible API
In addition, the server supports OpenAI-compatible APIs.

```python
import openai
client = openai.Client(
    base_url="http://127.0.0.1:30000/v1", api_key="EMPTY")

# Text completion
response = client.completions.create(
	model="default",
	prompt="The capital of France is",
	temperature=0,
	max_tokens=32,
)
print(response)

# Chat completion
response = client.chat.completions.create(
    model="default",
    messages=[
        {"role": "system", "content": "You are a helpful AI assistant"},
        {"role": "user", "content": "List 3 countries and their capitals."},
    ],
    temperature=0,
    max_tokens=64,
)
print(response)
202
203
204
205
206
207
208

# Text embedding
response = client.embeddings.create(
    model="default",
    input="How are you today",
)
print(response)
Ying Sheng's avatar
Ying Sheng committed
209
210
```

Lianmin Zheng's avatar
Lianmin Zheng committed
211
It supports streaming, vision, and almost all features of the Chat/Completions/Models/Batch endpoints specified by the [OpenAI API Reference](https://platform.openai.com/docs/api-reference/).
Ying Sheng's avatar
Ying Sheng committed
212
213

### Additional Server Arguments
214
- To enable multi-GPU tensor parallelism, add `--tp 2`. If it reports the error "peer access is not supported between these two devices", add `--enable-p2p-check` to the server launch command.
Ying Sheng's avatar
Ying Sheng committed
215
```
216
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 2
Ying Sheng's avatar
Ying Sheng committed
217
```
218
- To enable multi-GPU data parallelism, add `--dp 2`. Data parallelism is better for throughput if there is enough memory. It can also be used together with tensor parallelism. The following command uses 4 GPUs in total.
Ying Sheng's avatar
Ying Sheng committed
219
```
220
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --dp 2 --tp 2
Ying Sheng's avatar
Ying Sheng committed
221
```
222
- If you see out-of-memory errors during serving, try to reduce the memory usage of the KV cache pool by setting a smaller value of `--mem-fraction-static`. The default value is `0.9`.
Ying Sheng's avatar
Ying Sheng committed
223
```
224
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --mem-fraction-static 0.7
Ying Sheng's avatar
Ying Sheng committed
225
```
226
227
- See [hyperparameter_tuning.md](docs/en/hyperparameter_tuning.md) on tuning hyperparameters for better performance.
- If you see out-of-memory errors during prefill for long prompts, try to set a smaller chunked prefill size.
Ying Sheng's avatar
Ying Sheng committed
228
```
229
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --chunked-prefill-size 4096
Ying Sheng's avatar
Ying Sheng committed
230
```
231
- To enable torch.compile acceleration, add `--enable-torch-compile`. It accelerates small models on small batch sizes.
Lianmin Zheng's avatar
Lianmin Zheng committed
232
- To enable torchao quantization, add `--torchao-config int4wo-128`. It supports various quantization strategies.
233
234
235
- To enable fp8 weight quantization, add `--quantization fp8` on a fp16 checkpoint or directly load a fp8 checkpoint without specifying any arguments.
- To enable fp8 kv cache quantization, add `--kv-cache-dtype fp8_e5m2`.
- If the model does not have a chat template in the Hugging Face tokenizer, you can specify a [custom chat template](docs/en/custom_chat_template.md).
236
- To run tensor parallelism on multiple nodes, add `--nnodes 2`. If you have two nodes with two GPUs on each node and want to run TP=4, let `sgl-dev-0` be the hostname of the first node and `50000` be an available port, you can use the following commands. If you meet deadlock, please try to add `--disable-cuda-graph`
Ying Sheng's avatar
Ying Sheng committed
237
238
```
# Node 0
239
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-0:50000 --nnodes 2 --node-rank 0
Ying Sheng's avatar
Ying Sheng committed
240
241

# Node 1
242
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --tp 4 --nccl-init sgl-dev-0:50000 --nnodes 2 --node-rank 1
Ying Sheng's avatar
Ying Sheng committed
243
```
Lianmin Zheng's avatar
Lianmin Zheng committed
244
 
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
### Engine Without HTTP Server

We also provide an inference engine **without a HTTP server**. For example,

```python
import sglang as sgl


def main():
    prompts = [
        "Hello, my name is",
        "The president of the United States is",
        "The capital of France is",
        "The future of AI is",
    ]
    sampling_params = {"temperature": 0.8, "top_p": 0.95}
    llm = sgl.Engine(model_path="meta-llama/Meta-Llama-3.1-8B-Instruct")

    outputs = llm.generate(prompts, sampling_params)
    for prompt, output in zip(prompts, outputs):
        print("===============================")
        print(f"Prompt: {prompt}\nGenerated text: {output['text']}")

if __name__ == "__main__":
    main()
```

This can be used for:

1. **Offline Batch Inference**
2. **Building Custom Servers**

You can view the full example [here](https://github.com/sgl-project/sglang/tree/main/examples/runtime/engine)

Ying Sheng's avatar
Ying Sheng committed
279
280
### Supported Models

281
**Generative Models**
282
- Llama / Llama 2 / Llama 3 / Llama 3.1
283
- Mistral / Mixtral / Mistral NeMo
Ying Sheng's avatar
Ying Sheng committed
284
285
- Gemma / Gemma 2
- Qwen / Qwen 2 / Qwen 2 MoE
286
- DeepSeek / DeepSeek 2
Lianmin Zheng's avatar
Lianmin Zheng committed
287
- OLMoE
288
- [LLaVA-OneVision](https://llava-vl.github.io/blog/2024-08-05-llava-onevision/)
289
  - `python3 -m sglang.launch_server --model-path lmms-lab/llava-onevision-qwen2-7b-ov --port=30000 --chat-template=chatml-llava`
290
  - `python3 -m sglang.launch_server --model-path lmms-lab/llava-onevision-qwen2-72b-ov --port=30000 --tp-size=8 --chat-template=chatml-llava`
291
292
293
294
295
  - Query the server with the [OpenAI Vision API](https://platform.openai.com/docs/guides/vision). See examples at [test/srt/test_vision_openai_server.py](test/srt/test_vision_openai_server.py)
- LLaVA 1.5 / 1.6 / NeXT
  - `python -m sglang.launch_server --model-path lmms-lab/llama3-llava-next-8b --port=30000 --tp-size=1 --chat-template=llava_llama_3`
  - `python -m sglang.launch_server --model-path lmms-lab/llava-next-72b --port=30000 --tp-size=8 --chat-template=chatml-llava`
  - Query the server with the [OpenAI Vision API](https://platform.openai.com/docs/guides/vision). See examples at [test/srt/test_vision_openai_server.py](test/srt/test_vision_openai_server.py)
Ying Sheng's avatar
Ying Sheng committed
296
297
298
299
300
301
302
- Yi-VL
- StableLM
- Command-R
- DBRX
- Grok
- ChatGLM
- InternLM 2
303
- Exaone 3
Vectory's avatar
Vectory committed
304
- BaiChuan2
William's avatar
William committed
305
- MiniCPM / MiniCPM 3
306
- XVERSE / XVERSE MoE
307
- SmolLM
William's avatar
William committed
308

309
310
311
312
313
314
**Embedding Models**

- e5-mistral
- gte-Qwen2
  - `python -m sglang.launch_server --model-path Alibaba-NLP/gte-Qwen2-7B-instruct --is-embedding`

315
Instructions for supporting a new model are [here](docs/en/model_support.md).
Ying Sheng's avatar
Ying Sheng committed
316

Lianmin Zheng's avatar
Lianmin Zheng committed
317
#### Use Models From ModelScope
318
<details>
Dr. Artificial曾小健's avatar
Dr. Artificial曾小健 committed
319
<summary>More</summary>
320

Lianmin Zheng's avatar
Lianmin Zheng committed
321
To use a model from [ModelScope](https://www.modelscope.cn), set the environment variable SGLANG_USE_MODELSCOPE.
Lianmin Zheng's avatar
Lianmin Zheng committed
322
323
324
325
326
327
```
export SGLANG_USE_MODELSCOPE=true
```
Launch [Qwen2-7B-Instruct](https://www.modelscope.cn/models/qwen/qwen2-7b-instruct) Server
```
SGLANG_USE_MODELSCOPE=true python -m sglang.launch_server --model-path qwen/Qwen2-7B-Instruct --port 30000
Lianmin Zheng's avatar
Lianmin Zheng committed
328
```
329
330
331
332
333
334
335
336
337
338
339

Or start it by docker.
```bash
docker run --gpus all \
    -p 30000:30000 \
    -v ~/.cache/modelscope:/root/.cache/modelscope \
    --env "SGLANG_USE_MODELSCOPE=true" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server --model-path Qwen/Qwen2.5-7B-Instruct --host 0.0.0.0 --port 30000
```
Lianmin Zheng's avatar
Lianmin Zheng committed
340
  
341
</details>
Lianmin Zheng's avatar
Lianmin Zheng committed
342
343

#### Run Llama 3.1 405B
Lianmin Zheng's avatar
Lianmin Zheng committed
344
<details>
Dr. Artificial曾小健's avatar
Dr. Artificial曾小健 committed
345
<summary>More</summary>
Ying Sheng's avatar
Ying Sheng committed
346
347

```bash
348
# Run 405B (fp8) on a single node
Ying Sheng's avatar
Ying Sheng committed
349
350
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-405B-Instruct-FP8 --tp 8

351
352
353
# Run 405B (fp16) on two nodes
## on the first node, replace the `172.16.4.52:20000` with your own first node ip address and port
GLOO_SOCKET_IFNAME=eth0 python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-405B-Instruct --tp 16 --nccl-init-addr 172.16.4.52:20000 --nnodes 2 --node-rank 0 --disable-cuda-graph
Ying Sheng's avatar
Ying Sheng committed
354

355
356
## on the first node, replace the `172.16.4.52:20000` with your own first node ip address and port
GLOO_SOCKET_IFNAME=eth0 python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-405B-Instruct --tp 16 --nccl-init-addr 172.16.4.52:20000 --nnodes 2 --node-rank 1 --disable-cuda-graph
Ying Sheng's avatar
Ying Sheng committed
357
358
```

Lianmin Zheng's avatar
Lianmin Zheng committed
359
360
</details>

Ying Sheng's avatar
Ying Sheng committed
361
362
### Benchmark Performance

Lianmin Zheng's avatar
Lianmin Zheng committed
363
364
365
- Benchmark a single static batch by running the following command without launching a server. The arguments are the same as for `launch_server.py`.
  Note that this is not a dynamic batching server, so it may run out of memory for a batch size that a real server can handle.
  A real server truncates the prefill into several batches, while this unit test does not. For accurate large batch testing, please use `sglang.bench_serving` instead.
Ying Sheng's avatar
Ying Sheng committed
366
  ```
367
  python -m sglang.bench_latency --model-path meta-llama/Meta-Llama-3-8B-Instruct --batch 32 --input-len 256 --output-len 32
Ying Sheng's avatar
Ying Sheng committed
368
369
370
  ```
- Benchmark online serving. Launch a server first and run the following command.
  ```
371
  python3 -m sglang.bench_serving --backend sglang --num-prompt 10
Ying Sheng's avatar
Ying Sheng committed
372
373
  ```

Ying Sheng's avatar
Ying Sheng committed
374
## Frontend: Structured Generation Language (SGLang)
375
The frontend language can be used with local models or API models. It is an alternative to the OpenAI API. You may found it easier to use for complex prompting workflow.
Ying Sheng's avatar
Ying Sheng committed
376
377

### Quick Start
378
The example below shows how to use sglang to answer a multi-turn question.
Lianmin Zheng's avatar
Lianmin Zheng committed
379

Ying Sheng's avatar
Ying Sheng committed
380
#### Using Local Models
381
First, launch a server with
Lianmin Zheng's avatar
Lianmin Zheng committed
382
```
Ying Sheng's avatar
Ying Sheng committed
383
python -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000
Lianmin Zheng's avatar
Lianmin Zheng committed
384
385
```

386
387
Then, connect to the server and answer a multi-turn question.

Lianmin Zheng's avatar
Lianmin Zheng committed
388
```python
389
from sglang import function, system, user, assistant, gen, set_default_backend, RuntimeEndpoint
Lianmin Zheng's avatar
Lianmin Zheng committed
390
391
392
393
394
395
396
397
398

@function
def multi_turn_question(s, question_1, question_2):
    s += system("You are a helpful assistant.")
    s += user(question_1)
    s += assistant(gen("answer_1", max_tokens=256))
    s += user(question_2)
    s += assistant(gen("answer_2", max_tokens=256))

399
set_default_backend(RuntimeEndpoint("http://localhost:30000"))
Lianmin Zheng's avatar
Lianmin Zheng committed
400
401
402
403
404
405
406
407

state = multi_turn_question.run(
    question_1="What is the capital of the United States?",
    question_2="List two local attractions.",
)

for m in state.messages():
    print(m["role"], ":", m["content"])
408
409

print(state["answer_1"])
Lianmin Zheng's avatar
Lianmin Zheng committed
410
411
```

Ying Sheng's avatar
Ying Sheng committed
412
#### Using OpenAI Models
413
Set the OpenAI API Key
Lianmin Zheng's avatar
Lianmin Zheng committed
414
```
415
export OPENAI_API_KEY=sk-******
Lianmin Zheng's avatar
Lianmin Zheng committed
416
417
```

418
Then, answer a multi-turn question.
Lianmin Zheng's avatar
Lianmin Zheng committed
419
```python
420
from sglang import function, system, user, assistant, gen, set_default_backend, OpenAI
Lianmin Zheng's avatar
Lianmin Zheng committed
421
422
423
424
425
426
427
428
429

@function
def multi_turn_question(s, question_1, question_2):
    s += system("You are a helpful assistant.")
    s += user(question_1)
    s += assistant(gen("answer_1", max_tokens=256))
    s += user(question_2)
    s += assistant(gen("answer_2", max_tokens=256))

430
set_default_backend(OpenAI("gpt-3.5-turbo"))
Lianmin Zheng's avatar
Lianmin Zheng committed
431
432
433
434
435
436
437
438

state = multi_turn_question.run(
    question_1="What is the capital of the United States?",
    question_2="List two local attractions.",
)

for m in state.messages():
    print(m["role"], ":", m["content"])
439
440

print(state["answer_1"])
Lianmin Zheng's avatar
Lianmin Zheng committed
441
442
```

Ying Sheng's avatar
Ying Sheng committed
443
#### More Examples
Lianmin Zheng's avatar
Lianmin Zheng committed
444

445
Anthropic and VertexAI (Gemini) models are also supported.
Byron Hsu's avatar
Byron Hsu committed
446
You can find more examples at [examples/quick_start](examples/frontend_language/quick_start).
Lianmin Zheng's avatar
Lianmin Zheng committed
447

Ying Sheng's avatar
Ying Sheng committed
448
### Language Feature
Lianmin Zheng's avatar
Lianmin Zheng committed
449
450
451
452
453
To begin with, import sglang.
```python
import sglang as sgl
```

Lianmin Zheng's avatar
Lianmin Zheng committed
454
`sglang` provides some simple primitives such as `gen`, `select`, `fork`, `image`.
Lianmin Zheng's avatar
Lianmin Zheng committed
455
456
You can implement your prompt flow in a function decorated by `sgl.function`.
You can then invoke the function with `run` or `run_batch`.
457
The system will manage the state, chat template, parallelism and batching for you.
Lianmin Zheng's avatar
Lianmin Zheng committed
458

459
The complete code for the examples below can be found at [readme_examples.py](examples/frontend_language/usage/readme_examples.py)
460

Ying Sheng's avatar
Ying Sheng committed
461
#### Control Flow
Lianmin Zheng's avatar
Lianmin Zheng committed
462
463
You can use any Python code within the function body, including control flow, nested function calls, and external libraries.

Lianmin Zheng's avatar
Lianmin Zheng committed
464
465
```python
@sgl.function
466
467
468
def tool_use(s, question):
    s += "To answer this question: " + question + ". "
    s += "I need to use a " + sgl.gen("tool", choices=["calculator", "search engine"]) + ". "
Lianmin Zheng's avatar
Lianmin Zheng committed
469
470
471

    if s["tool"] == "calculator":
        s += "The math expression is" + sgl.gen("expression")
472
473
    elif s["tool"] == "search engine":
        s += "The key word to search is" + sgl.gen("word")
Lianmin Zheng's avatar
Lianmin Zheng committed
474
```
Lianmin Zheng's avatar
Lianmin Zheng committed
475

Ying Sheng's avatar
Ying Sheng committed
476
#### Parallelism
Lianmin Zheng's avatar
Lianmin Zheng committed
477
478
479
Use `fork` to launch parallel prompts.
Because `sgl.gen` is non-blocking, the for loop below issues two generation calls in parallel.

Lianmin Zheng's avatar
Lianmin Zheng committed
480
481
482
483
484
485
486
487
```python
@sgl.function
def tip_suggestion(s):
    s += (
        "Here are two tips for staying healthy: "
        "1. Balanced Diet. 2. Regular Exercise.\n\n"
    )

Lianmin Zheng's avatar
Lianmin Zheng committed
488
    forks = s.fork(2)
Lianmin Zheng's avatar
Lianmin Zheng committed
489
490
491
492
493
494
495
496
    for i, f in enumerate(forks):
        f += f"Now, expand tip {i+1} into a paragraph:\n"
        f += sgl.gen(f"detailed_tip", max_tokens=256, stop="\n\n")

    s += "Tip 1:" + forks[0]["detailed_tip"] + "\n"
    s += "Tip 2:" + forks[1]["detailed_tip"] + "\n"
    s += "In summary" + sgl.gen("summary")
```
Lianmin Zheng's avatar
Lianmin Zheng committed
497

Lianmin Zheng's avatar
Lianmin Zheng committed
498
#### Multi-Modality
Lianmin Zheng's avatar
Lianmin Zheng committed
499
500
Use `sgl.image` to pass an image as input.

Lianmin Zheng's avatar
Lianmin Zheng committed
501
502
```python
@sgl.function
Lianmin Zheng's avatar
Lianmin Zheng committed
503
def image_qa(s, image_file, question):
Lianmin Zheng's avatar
Lianmin Zheng committed
504
    s += sgl.user(sgl.image(image_file) + question)
Lianmin Zheng's avatar
Lianmin Zheng committed
505
    s += sgl.assistant(sgl.gen("answer", max_tokens=256)
Lianmin Zheng's avatar
Lianmin Zheng committed
506
507
```

508
See also [srt_example_llava.py](examples/frontend_language/quick_start/local_example_llava_next.py).
509

Ying Sheng's avatar
Ying Sheng committed
510
#### Constrained Decoding
511
512
Use `regex` to specify a regular expression as a decoding constraint.
This is only supported for local models.
Lianmin Zheng's avatar
Lianmin Zheng committed
513

Lianmin Zheng's avatar
Lianmin Zheng committed
514
```python
Lianmin Zheng's avatar
Lianmin Zheng committed
515
@sgl.function
Lianmin Zheng's avatar
Lianmin Zheng committed
516
517
def regular_expression_gen(s):
    s += "Q: What is the IP address of the Google DNS servers?\n"
Lianmin Zheng's avatar
Lianmin Zheng committed
518
    s += "A: " + sgl.gen(
Lianmin Zheng's avatar
Lianmin Zheng committed
519
520
521
522
523
        "answer",
        temperature=0,
        regex=r"((25[0-5]|2[0-4]\d|[01]?\d\d?).){3}(25[0-5]|2[0-4]\d|[01]?\d\d?)",
    )
```
Lianmin Zheng's avatar
Lianmin Zheng committed
524

Ying Sheng's avatar
Ying Sheng committed
525
#### JSON Decoding
Lianmin Zheng's avatar
Lianmin Zheng committed
526
Use `regex` to specify a JSON schema with a regular expression.
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547

```python
character_regex = (
    r"""\{\n"""
    + r"""    "name": "[\w\d\s]{1,16}",\n"""
    + r"""    "house": "(Gryffindor|Slytherin|Ravenclaw|Hufflepuff)",\n"""
    + r"""    "blood status": "(Pure-blood|Half-blood|Muggle-born)",\n"""
    + r"""    "occupation": "(student|teacher|auror|ministry of magic|death eater|order of the phoenix)",\n"""
    + r"""    "wand": \{\n"""
    + r"""        "wood": "[\w\d\s]{1,16}",\n"""
    + r"""        "core": "[\w\d\s]{1,16}",\n"""
    + r"""        "length": [0-9]{1,2}\.[0-9]{0,2}\n"""
    + r"""    \},\n"""
    + r"""    "alive": "(Alive|Deceased)",\n"""
    + r"""    "patronus": "[\w\d\s]{1,16}",\n"""
    + r"""    "bogart": "[\w\d\s]{1,16}"\n"""
    + r"""\}"""
)

@sgl.function
def character_gen(s, name):
Lianmin Zheng's avatar
Lianmin Zheng committed
548
    s += name + " is a character in Harry Potter. Please fill in the following information about this character.\n"
549
550
551
    s += sgl.gen("json_output", max_tokens=256, regex=character_regex)
```

552
See also [json_decode.py](examples/frontend_language/usage/json_decode.py) for an additional example of specifying formats with Pydantic models.
553

Ying Sheng's avatar
Ying Sheng committed
554
#### Batching
Lianmin Zheng's avatar
Lianmin Zheng committed
555
556
Use `run_batch` to run a batch of requests with continuous batching.

Lianmin Zheng's avatar
Lianmin Zheng committed
557
558
559
560
561
562
563
564
565
566
567
568
```python
@sgl.function
def text_qa(s, question):
    s += "Q: " + question + "\n"
    s += "A:" + sgl.gen("answer", stop="\n")

states = text_qa.run_batch(
    [
        {"question": "What is the capital of the United Kingdom?"},
        {"question": "What is the capital of France?"},
        {"question": "What is the capital of Japan?"},
    ],
Lianmin Zheng's avatar
Lianmin Zheng committed
569
    progress_bar=True
Lianmin Zheng's avatar
Lianmin Zheng committed
570
571
)
```
Lianmin Zheng's avatar
Lianmin Zheng committed
572

Ying Sheng's avatar
Ying Sheng committed
573
#### Streaming
Lianmin Zheng's avatar
Lianmin Zheng committed
574
575
Add `stream=True` to enable streaming.

Lianmin Zheng's avatar
Lianmin Zheng committed
576
577
578
579
580
581
```python
@sgl.function
def text_qa(s, question):
    s += "Q: " + question + "\n"
    s += "A:" + sgl.gen("answer", stop="\n")

582
state = text_qa.run(
Lianmin Zheng's avatar
Lianmin Zheng committed
583
    question="What is the capital of France?",
Lianmin Zheng's avatar
Lianmin Zheng committed
584
585
586
    temperature=0.1,
    stream=True
)
Lianmin Zheng's avatar
Lianmin Zheng committed
587

Lianmin Zheng's avatar
Lianmin Zheng committed
588
589
590
for out in state.text_iter():
    print(out, end="", flush=True)
```
Lianmin Zheng's avatar
Lianmin Zheng committed
591

592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
#### Roles

Use `sgl.system``sgl.user` and `sgl.assistant` to set roles when using Chat models. You can also define more complex role prompts using begin and end tokens.

```python
@sgl.function
def chat_example(s):
    s += sgl.system("You are a helpful assistant.")
    # Same as: s += s.system("You are a helpful assistant.")

    with s.user():
        s += "Question: What is the capital of France?"

    s += sgl.assistant_begin()
    s += "Answer: " + sgl.gen(max_tokens=100, stop="\n")
    s += sgl.assistant_end()
```

Ying Sheng's avatar
Ying Sheng committed
610
#### Tips and Implementation Details
611
612
- The `choices` argument in `sgl.gen` is implemented by computing the [token-length normalized log probabilities](https://blog.eleuther.ai/multiple-choice-normalization/) of all choices and selecting the one with the highest probability.
- The `regex` argument in `sgl.gen` is implemented through autoregressive decoding with logit bias masking, according to the constraints set by the regex. It is compatible with `temperature=0` and `temperature != 0`.
Lianmin Zheng's avatar
Lianmin Zheng committed
613

Ying Sheng's avatar
Ying Sheng committed
614
615
616
## Benchmark And Performance
![8b_throughput](https://lmsys.org/images/blog/sglang_llama3/8b_throughput.svg)
![70b_fp8_throughput](https://lmsys.org/images/blog/sglang_llama3/70b_fp8_throughput.svg)
Lianmin Zheng's avatar
Lianmin Zheng committed
617

Ying Sheng's avatar
Ying Sheng committed
618
Learn more at this [blog](https://lmsys.org/blog/2024-07-25-sglang-llama3/).
Lianmin Zheng's avatar
Lianmin Zheng committed
619

Lianmin Zheng's avatar
Lianmin Zheng committed
620
## Roadmap
621
[Development Roadmap (2024 Q4)](https://github.com/sgl-project/sglang/issues/1487)
Lianmin Zheng's avatar
Lianmin Zheng committed
622
623

## Citation And Acknowledgment
Ying Sheng's avatar
Ying Sheng committed
624
625
Please cite our paper, [SGLang: Efficient Execution of Structured Language Model Programs](https://arxiv.org/abs/2312.07104), if you find the project useful.
We also learned from the design and reused code from the following projects: [Guidance](https://github.com/guidance-ai/guidance), [vLLM](https://github.com/vllm-project/vllm), [LightLLM](https://github.com/ModelTC/lightllm), [FlashInfer](https://github.com/flashinfer-ai/flashinfer), [Outlines](https://github.com/outlines-dev/outlines), and [LMQL](https://github.com/eth-sri/lmql).
626
627
628
629
630
631
632
633



<p align="center">
  <a href="#sglangtop" target="_blank">
  <bold>Back To Top </bold>
  </a>
</p>