Query VLM with Offline Engine

Query VLM with Offline Engine#

This tutorial demonstrates how to use SGLang’s offline Engine API to query VLMs. We will demonstrate usage with Qwen2.5-VL and Llama 4. This section demonstrates three different calling approaches:

  1. Basic Call: Directly pass images and text.

  2. Processor Output: Use HuggingFace processor for data preprocessing.

  3. Precomputed Embeddings: Pre-calculate image features to improve inference efficiency.

Understanding the Three Input Formats#

SGLang supports three ways to pass visual data, each optimized for different scenarios:

1. Raw Images - Simplest approach#

  • Pass PIL Images, file paths, URLs, or base64 strings directly

  • SGLang handles all preprocessing automatically

  • Best for: Quick prototyping, simple applications

2. Processor Output - For custom preprocessing#

  • Pre-process images with HuggingFace processor

  • Pass the complete processor output dict with format: "processor_output"

  • Best for: Custom image transformations, integration with existing pipelines

  • Requirement: Must use input_ids instead of text prompt

3. Precomputed Embeddings - For maximum performance#

  • Pre-calculate visual embeddings using the vision encoder

  • Pass embeddings with format: "precomputed_embedding"

  • Best for: Repeated queries on same images, caching, high-throughput serving

  • Performance gain: Avoids redundant vision encoder computation (30-50% speedup)

Key Rule: Within a single request, use only one format for all images. Don’t mix formats.

The examples below demonstrate all three approaches with both Qwen2.5-VL and Llama 4 models.

Querying Qwen2.5-VL Model#

[1]:

import nest_asyncio

nest_asyncio.apply()

import sglang.test.doc_patch  # noqa: F401

model_path = "Qwen/Qwen2.5-VL-3B-Instruct"
chat_template = "qwen2-vl"
example_image_url = "https://raw.githubusercontent.com/sgl-project/sglang/main/examples/assets/example_image.png"

[2]:

from io import BytesIO
import requests
from PIL import Image

from sglang.srt.parser.conversation import chat_templates

image = Image.open(BytesIO(requests.get(example_image_url).content))

conv = chat_templates[chat_template].copy()
conv.append_message(conv.roles[0], f"What's shown here: {conv.image_token}?")
conv.append_message(conv.roles[1], "")
conv.image_data = [image]

print("Generated prompt text:")
print(conv.get_prompt())
print(f"\nImage size: {image.size}")
image

Generated prompt text:
<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
What's shown here: <|vision_start|><|image_pad|><|vision_end|>?<|im_end|>
<|im_start|>assistant


Image size: (570, 380)

[2]:
../_images/advanced_features_vlm_query_4_1.png

Basic Offline Engine API Call#

[3]:

from sglang import Engine

llm = Engine(model_path=model_path, chat_template=chat_template, log_level="warning")

[transformers] The `use_fast` parameter is deprecated and will be removed in a future version. Use `backend="torchvision"` instead of `use_fast=True`, or `backend="pil"` instead of `use_fast=False`.
[2026-05-14 09:40:20] The `use_fast` parameter is deprecated and will be removed in a future version. Use `backend="torchvision"` instead of `use_fast=True`, or `backend="pil"` instead of `use_fast=False`.
[transformers] The `use_fast` parameter is deprecated and will be removed in a future version. Use `backend="torchvision"` instead of `use_fast=True`, or `backend="pil"` instead of `use_fast=False`.
[2026-05-14 09:40:24] The `use_fast` parameter is deprecated and will be removed in a future version. Use `backend="torchvision"` instead of `use_fast=True`, or `backend="pil"` instead of `use_fast=False`.
Multi-thread loading shards: 100% Completed | 2/2 [00:01<00:00,  1.20it/s]

[4]:

out = llm.generate(prompt=conv.get_prompt(), image_data=[image])
print("Model response:")
print(out["text"])

2026-05-14 09:40:32,848 - CUTE_DSL - WARNING - [handle_import_error] - Unexpected error during package walk: cutlass.cute.experimental
[2026-05-14 09:40:32] Unexpected error during package walk: cutlass.cute.experimental

Model response:
The image shows a unique scene of a man hanging waves or fishing nets on the back of yellow taxis in New York City. The practice is known as " waving taxis" and was once a popularSzczodrej tactic employed across the United States in the 1970s.

Call with Processor Output#

Using a HuggingFace processor to preprocess text and images, and passing the processor_output directly into Engine.generate.

[5]:

from transformers import AutoProcessor

processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
processor_output = processor(
    images=[image], text=conv.get_prompt(), return_tensors="pt"
)

out = llm.generate(
    input_ids=processor_output["input_ids"][0].detach().cpu().tolist(),
    image_data=[dict(processor_output, format="processor_output")],
)
print("Response using processor output:")
print(out["text"])

Response using processor output:
This image depicts a scene in what appears to be a busy urban area, likely New York City, given the bright yellow taxis that are iconic in the city. A person in a yellow shirt and glasses is pushing a folding table loaded with clothes on a city street. The flags visible in the background include the New York City flag, suggesting the presence of the Holland Tunnel to the east and the Brooklyn Navy Yard to the west, both lined with the city flag as a civic symbol.

The taxi in the foreground is a typical New York City yellow cab, and the street is bustling with activity. The food truck on the right (towards the

Call with Precomputed Embeddings#

You can pre-calculate image features to avoid repeated visual encoding processes.

[6]:

from transformers import AutoProcessor
from transformers import Qwen2_5_VLForConditionalGeneration

processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(model_path).eval()
vision = model.model.visual.cuda()

[7]:

processor_output = processor(
    images=[image], text=conv.get_prompt(), return_tensors="pt"
)

input_ids = processor_output["input_ids"][0].detach().cpu().tolist()

precomputed_embeddings = vision(
    processor_output["pixel_values"].cuda(), processor_output["image_grid_thw"].cuda()
)
precomputed_embeddings = precomputed_embeddings.pooler_output

multi_modal_item = dict(
    processor_output,
    format="precomputed_embedding",
    feature=precomputed_embeddings,
)

out = llm.generate(input_ids=input_ids, image_data=[multi_modal_item])
print("Response using precomputed embeddings:")
print(out["text"])

llm.shutdown()

Response using precomputed embeddings:
The image shows two identical blue duffel bags hanging from a clothes rack or a similar makeshift display in front of a yellow taxi cab. The blue duffel bag hanging on the right side is the back view of it, revealing the open lid. This setup is likely part of a street marketing or performance art demonstration, where the bags might be intended to paint the colors blue.

There's an unusual piece of urban event or art that employs bright colors (blue in the image) to catch passersby's attention and perhaps activate a conversation or gathering.

Querying Llama 4 Vision Model#

model_path = "meta-llama/Llama-4-Scout-17B-16E-Instruct"
chat_template = "llama-4"

from io import BytesIO
import requests
from PIL import Image

from sglang.srt.parser.conversation import chat_templates

# Download the same example image
image = Image.open(BytesIO(requests.get(example_image_url).content))

conv = chat_templates[chat_template].copy()
conv.append_message(conv.roles[0], f"What's shown here: {conv.image_token}?")
conv.append_message(conv.roles[1], "")
conv.image_data = [image]

print("Llama 4 generated prompt text:")
print(conv.get_prompt())
print(f"Image size: {image.size}")

image

Llama 4 Basic Call#

Llama 4 requires more computational resources, so it’s configured with multi-GPU parallelism (tp_size=4) and larger context length.

llm = Engine(
    model_path=model_path,
    enable_multimodal=True,
    attention_backend="fa3",
    tp_size=4,
    context_length=65536,
)

out = llm.generate(prompt=conv.get_prompt(), image_data=[image])
print("Llama 4 response:")
print(out["text"])

Call with Processor Output#

Using HuggingFace processor to preprocess data can reduce computational overhead during inference.

from transformers import AutoProcessor

processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
processor_output = processor(
    images=[image], text=conv.get_prompt(), return_tensors="pt"
)

out = llm.generate(
    input_ids=processor_output["input_ids"][0].detach().cpu().tolist(),
    image_data=[dict(processor_output, format="processor_output")],
)
print("Response using processor output:")
print(out)

Call with Precomputed Embeddings#

from transformers import AutoProcessor
from transformers import Llama4ForConditionalGeneration

processor = AutoProcessor.from_pretrained(model_path, use_fast=True)
model = Llama4ForConditionalGeneration.from_pretrained(
    model_path, torch_dtype="auto"
).eval()

vision = model.vision_model.cuda()
multi_modal_projector = model.multi_modal_projector.cuda()

print(f'Image pixel values shape: {processor_output["pixel_values"].shape}')
input_ids = processor_output["input_ids"][0].detach().cpu().tolist()

# Process image through vision encoder
image_outputs = vision(
    processor_output["pixel_values"].to("cuda"),
    aspect_ratio_ids=processor_output["aspect_ratio_ids"].to("cuda"),
    aspect_ratio_mask=processor_output["aspect_ratio_mask"].to("cuda"),
    output_hidden_states=False
)
image_features = image_outputs.last_hidden_state

# Flatten image features and pass through multimodal projector
vision_flat = image_features.view(-1, image_features.size(-1))
precomputed_embeddings = multi_modal_projector(vision_flat)

# Build precomputed embedding data item
mm_item = dict(
    processor_output,
    format="precomputed_embedding",
    feature=precomputed_embeddings
)

# Use precomputed embeddings for efficient inference
out = llm.generate(input_ids=input_ids, image_data=[mm_item])
print("Llama 4 precomputed embedding response:")
print(out["text"])
Contents