Instructions to use wangkanai/qwen3-vl-8b-thinking with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use wangkanai/qwen3-vl-8b-thinking with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="wangkanai/qwen3-vl-8b-thinking")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("wangkanai/qwen3-vl-8b-thinking", dtype="auto")

llama-cpp-python

How to use wangkanai/qwen3-vl-8b-thinking with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="wangkanai/qwen3-vl-8b-thinking",
	filename="qwen3-vl-8b-thinking-abliterated-f16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": [
				{
					"type": "text",
					"text": "Describe this image in one sentence."
				},
				{
					"type": "image_url",
					"image_url": {
						"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
					}
				}
			]
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use wangkanai/qwen3-vl-8b-thinking with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf wangkanai/qwen3-vl-8b-thinking:F16
# Run inference directly in the terminal:
llama-cli -hf wangkanai/qwen3-vl-8b-thinking:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf wangkanai/qwen3-vl-8b-thinking:F16
# Run inference directly in the terminal:
llama-cli -hf wangkanai/qwen3-vl-8b-thinking:F16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf wangkanai/qwen3-vl-8b-thinking:F16
# Run inference directly in the terminal:
./llama-cli -hf wangkanai/qwen3-vl-8b-thinking:F16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf wangkanai/qwen3-vl-8b-thinking:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf wangkanai/qwen3-vl-8b-thinking:F16

Use Docker

docker model run hf.co/wangkanai/qwen3-vl-8b-thinking:F16

LM Studio
Jan

vLLM

How to use wangkanai/qwen3-vl-8b-thinking with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "wangkanai/qwen3-vl-8b-thinking"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "wangkanai/qwen3-vl-8b-thinking",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/wangkanai/qwen3-vl-8b-thinking:F16

SGLang

How to use wangkanai/qwen3-vl-8b-thinking with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "wangkanai/qwen3-vl-8b-thinking" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "wangkanai/qwen3-vl-8b-thinking",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "wangkanai/qwen3-vl-8b-thinking" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "wangkanai/qwen3-vl-8b-thinking",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Ollama
How to use wangkanai/qwen3-vl-8b-thinking with Ollama:
```
ollama run hf.co/wangkanai/qwen3-vl-8b-thinking:F16
```

Unsloth Studio new

How to use wangkanai/qwen3-vl-8b-thinking with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for wangkanai/qwen3-vl-8b-thinking to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for wangkanai/qwen3-vl-8b-thinking to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for wangkanai/qwen3-vl-8b-thinking to start chatting

Pi new

How to use wangkanai/qwen3-vl-8b-thinking with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf wangkanai/qwen3-vl-8b-thinking:F16

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "wangkanai/qwen3-vl-8b-thinking:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use wangkanai/qwen3-vl-8b-thinking with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf wangkanai/qwen3-vl-8b-thinking:F16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default wangkanai/qwen3-vl-8b-thinking:F16

Run Hermes

hermes

Docker Model Runner
How to use wangkanai/qwen3-vl-8b-thinking with Docker Model Runner:
```
docker model run hf.co/wangkanai/qwen3-vl-8b-thinking:F16
```

Lemonade

How to use wangkanai/qwen3-vl-8b-thinking with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull wangkanai/qwen3-vl-8b-thinking:F16

Run and chat with the model

lemonade run user.qwen3-vl-8b-thinking-F16

List all available models

lemonade list

Qwen 3 VL 8B Thinking

Model Description

Qwen 3 VL 8B Thinking is a vision-language multimodal model with 8 billion parameters, part of the Qwen 3 family developed by Alibaba Cloud. This model combines vision and language understanding capabilities with extended "thinking" or reasoning capabilities for complex visual question answering and multimodal tasks.

Key Capabilities:

🖼️ Vision-Language Understanding: Process and understand images with natural language
🧠 Reasoning Capabilities: Extended thinking process for complex visual reasoning
💬 Multimodal Chat: Interactive conversations about images and visual content
🎯 Visual Question Answering: Answer questions about image content with detailed reasoning
📊 Scene Understanding: Comprehensive analysis of visual scenes and contexts

Repository Contents

⚠️ Note: This directory is currently being prepared for model files.

Expected model structure:

qwen3-vl-8b-thinking/
├── config.json                    # Model configuration
├── model.safetensors              # Main model weights (~16GB)
├── tokenizer.json                 # Tokenizer configuration
├── tokenizer_config.json          # Tokenizer settings
├── special_tokens_map.json        # Special tokens mapping
├── preprocessor_config.json       # Image preprocessor config
├── generation_config.json         # Generation parameters
└── README.md                      # This file

Expected Total Size: ~16-20 GB (FP16 precision)

Hardware Requirements

Minimum Requirements

VRAM: 20GB+ (RTX 4090, A5000, or better)
System RAM: 32GB recommended
Disk Space: 25GB free space
CUDA: 11.8 or higher recommended

Recommended Requirements

VRAM: 24GB+ (RTX 4090, A6000, A100)
System RAM: 64GB for optimal performance
Disk Space: 50GB for model + cache
CUDA: 12.0+ for best performance

Performance Estimates

FP16: ~20GB VRAM, fastest inference
8-bit quantization: ~10GB VRAM, good quality
4-bit quantization: ~6GB VRAM, acceptable quality

Usage Examples

Basic Usage with Transformers

from transformers import AutoModelForVision2Seq, AutoProcessor
from PIL import Image
import torch

# Load model and processor
model_path = "E:/huggingface/qwen3-vl-8b-thinking"
model = AutoModelForVision2Seq.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    device_map="auto"
)
processor = AutoProcessor.from_pretrained(model_path)

# Load and process image
image = Image.open("example.jpg")
prompt = "Describe this image in detail and explain what's happening."

# Prepare inputs
inputs = processor(
    text=prompt,
    images=image,
    return_tensors="pt"
).to(model.device)

# Generate response with thinking
outputs = model.generate(
    **inputs,
    max_new_tokens=512,
    do_sample=True,
    temperature=0.7,
    top_p=0.9
)

# Decode response
response = processor.decode(outputs[0], skip_special_tokens=True)
print(response)

Visual Question Answering

# Ask specific questions about images
questions = [
    "What objects are visible in this image?",
    "What is the main activity taking place?",
    "What might happen next in this scene?"
]

for question in questions:
    inputs = processor(
        text=question,
        images=image,
        return_tensors="pt"
    ).to(model.device)

    outputs = model.generate(**inputs, max_new_tokens=256)
    answer = processor.decode(outputs[0], skip_special_tokens=True)
    print(f"Q: {question}")
    print(f"A: {answer}\n")

Batch Processing Multiple Images

from pathlib import Path

# Process multiple images
image_dir = Path("images/")
images = [Image.open(img) for img in image_dir.glob("*.jpg")]
prompts = ["Analyze this image:"] * len(images)

# Batch processing
inputs = processor(
    text=prompts,
    images=images,
    return_tensors="pt",
    padding=True
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
responses = [processor.decode(out, skip_special_tokens=True) for out in outputs]

for img_path, response in zip(image_dir.glob("*.jpg"), responses):
    print(f"\n{img_path.name}:")
    print(response)

Memory-Efficient Loading (8-bit)

from transformers import BitsAndBytesConfig

# 8-bit quantization for lower VRAM usage
quantization_config = BitsAndBytesConfig(
    load_in_8bit=True,
    bnb_8bit_compute_dtype=torch.float16
)

model = AutoModelForVision2Seq.from_pretrained(
    model_path,
    quantization_config=quantization_config,
    device_map="auto"
)

# Use as normal - ~50% VRAM reduction

Model Specifications

Architecture

Base Architecture: Qwen 3 Vision-Language Transformer
Parameters: 8 billion
Vision Encoder: High-resolution vision transformer
Language Model: Qwen 3 8B language backbone
Context Length: Up to 8K tokens
Image Resolution: Dynamic resolution support (up to 1024x1024)

Precision and Format

Default Precision: FP16 (Float16)
Format: SafeTensors (secure, efficient)
Quantization Support: 8-bit, 4-bit via bitsandbytes
Framework: PyTorch with Transformers

Training Details

Base Model: Qwen 3 VL 8B
Special Training: Extended reasoning/thinking capabilities
Multimodal Alignment: Vision-language co-training
Optimization: Instruction-tuned for visual understanding

Performance Tips

Optimization Recommendations

Use Flash Attention 2 (if available):

model = AutoModelForVision2Seq.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    attn_implementation="flash_attention_2",
    device_map="auto"
)

Enable Compilation (PyTorch 2.0+):

model = torch.compile(model, mode="reduce-overhead")

Optimize Image Preprocessing:

# Resize large images before processing
from PIL import Image

def preprocess_image(img_path, max_size=1024):
    img = Image.open(img_path)
    if max(img.size) > max_size:
        img.thumbnail((max_size, max_size), Image.Resampling.LANCZOS)
    return img

Batch Similar-Sized Images: Group images by size for efficient batch processing
Use Lower Precision for Inference: FP16 or BF16 for speed, 8-bit for VRAM constraints

Memory Management

import torch
import gc

# Clear cache between batches
def clear_memory():
    gc.collect()
    torch.cuda.empty_cache()

# Use after processing batches
clear_memory()

License

This model is released under the Apache 2.0 License.

You are free to:

✅ Use commercially
✅ Modify and distribute
✅ Use privately
✅ Use for patents

Conditions:

📄 Include license and copyright notice
📝 State changes made to the code
🔓 Distribute under same license

See the Apache 2.0 License for full terms.

Citation

If you use this model in your research or applications, please cite:

@misc{qwen3-vl-8b-thinking,
  title={Qwen 3 VL 8B Thinking: Vision-Language Model with Reasoning},
  author={Qwen Team, Alibaba Cloud},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/Qwen/Qwen3-VL-8B-Thinking}}
}

Resources and Links

🏠 Official Website: https://qwenlm.github.io/
📚 Documentation: https://huggingface.co/docs/transformers/main/en/model_doc/qwen3
💬 Community: https://huggingface.co/Qwen
🐛 Issues: Report issues on the official Qwen GitHub repository
📖 Paper: Check Qwen technical reports for architecture details

Supported Tasks

Visual Question Answering (VQA): Answer questions about image content
Image Captioning: Generate detailed descriptions of images
Visual Reasoning: Complex reasoning about visual scenes
Multimodal Chat: Interactive conversations with image context
Scene Understanding: Comprehensive analysis of visual contexts
Object Recognition: Identify and describe objects in images

Model Limitations

Image resolution limits may affect fine detail recognition
Performance varies based on image quality and clarity
May require fine-tuning for domain-specific applications
Reasoning capabilities depend on prompt quality and structure
Computational requirements may limit deployment scenarios

Safety and Responsible Use

Review outputs for accuracy, especially in critical applications
Be aware of potential biases in visual understanding
Validate model responses for factual correctness
Use appropriate safety filters for production deployments
Consider privacy implications when processing images

Version: 1.0 Last Updated: 2025-11-05 Model Type: Vision-Language Multimodal Status: Ready for local deployment

Downloads last month: 23

GGUF

Model size

8B params

Architecture

qwen3vl

Hardware compatibility

16-bit

View +2 variants

Collection including wangkanai/qwen3-vl-8b-thinking

qwen3-vl

Collection

Qwen3 vision language • 9 items • Updated Nov 5, 2025 • 1