SGLang Deployment

Deploy Qwen3-VL with SGLang for efficient inference and serving

Overview

SGLang provides an alternative high-performance serving solution for Qwen3-VL models. You can start an SGLang server to serve models efficiently with an OpenAI-style API.

Starting the SGLang Server

Launch the SGLang server with the following command:

python -m sglang.launch_server \
   --model-path Qwen/Qwen3-VL-235B-A22B-Instruct \
   --host 0.0.0.0 \
   --port 22002 \
   --tp 4

Server Parameters

• --model-path: Path to the model (local path or HuggingFace model ID)
• --host: Server host address (default: 0.0.0.0)
• --port: Server port (default: 22002)
• --tp: Tensor parallelism size for multi-GPU deployment

Making Requests

Once the server is running, you can make requests using the OpenAI-compatible API.

Image Request Example

import time
from openai import OpenAI

client = OpenAI(
    api_key="EMPTY",
    base_url="http://127.0.0.1:22002/v1",
    timeout=3600
)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image_url",
                "image_url": {
                    "url": "https://ofasys-multimodal-wlcb-3-toshanghai.oss-accelerate.aliyuncs.com/wpf272043/keepme/image/receipt.png"
                }
            },
            {
                "type": "text",
                "text": "Read all the text in the image."
            }
        ]
    }
]

start = time.time()
response = client.chat.completions.create(
    model="Qwen/Qwen3-VL-235B-A22B-Instruct",
    messages=messages,
    max_tokens=2048
)
print(f"Response costs: {time.time() - start:.2f}s")
print(f"Generated text: {response.choices[0].message.content}")

Video Request Example

import time
from openai import OpenAI

client = OpenAI(
    api_key="EMPTY",
    base_url="http://127.0.0.1:22002/v1",
    timeout=3600
)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "video_url",
                "video_url": {
                    "url": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4"
                }
            },
            {
                "type": "text",
                "text": "How long is this video?"
            }
        ]
    }
]

start = time.time()
response = client.chat.completions.create(
    model="Qwen/Qwen3-VL-235B-A22B-Instruct",
    messages=messages,
    max_tokens=2048
)
print(f"Response costs: {time.time() - start:.2f}s")
print(f"Generated text: {response.choices[0].message.content}")

Offline Inference

You can also use SGLang for local offline inference without running a server:

import time
from PIL import Image
from sglang import Engine
from qwen_vl_utils import process_vision_info
from transformers import AutoProcessor, AutoConfig


if __name__ == "__main__":
    # TODO: change to your own checkpoint path
    checkpoint_path = "Qwen/Qwen3-VL-235B-A22B-Instruct"
    processor = AutoProcessor.from_pretrained(checkpoint_path)

    messages = [
        {
            "role": "user",
            "content": [
              {
                  "type": "image",
                  "image": "https://ofasys-multimodal-wlcb-3-toshanghai.oss-accelerate.aliyuncs.com/wpf272043/keepme/image/receipt.png",
              },
              {"type": "text", "text": "Read all the text in the image."},
            ],
        }
    ]

    text = processor.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )

    image_inputs, _ = process_vision_info(messages, image_patch_size=processor.image_processor.patch_size)

    llm = Engine(
        model_path=checkpoint_path,
        enable_multimodal=True,
        mem_fraction_static=0.8,
        tp_size=4,
        attention_backend="fa3",
        context_length=10240,
        disable_cuda_graph=True,
    )

    start = time.time()
    sampling_params = {"max_new_tokens": 1024}
    response = llm.generate(prompt=text, image_data=image_inputs, sampling_params=sampling_params)
    print(f"Response costs: {time.time() - start:.2f}s")
    print(f"Generated text: {response['text']}")

Engine Configuration

The SGLang Engine supports various configuration options:

• model_path: Path to the model checkpoint
• enable_multimodal: Enable multimodal support (required for Qwen3-VL)
• mem_fraction_static: GPU memory fraction for static allocation (0.0-1.0)
• tp_size: Tensor parallelism size for multi-GPU deployment
• attention_backend: Attention implementation (e.g., “fa3” for FlashAttention 3)
• context_length: Maximum context length
• disable_cuda_graph: Disable CUDA graph optimization (may be needed for stability)

Next Steps

• Compare with vLLM deployment
• Try the Docker deployment for quick setup
• Learn about the DashScope API service