How to use from
vLLM
Install from pip and serve model
# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "chaser2/VIG-Qwen3-VL-Thinking-8B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "chaser2/VIG-Qwen3-VL-Thinking-8B",
		"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/chaser2/VIG-Qwen3-VL-Thinking-8B
Quick Links

VIG Qwen3-VL-Thinking 8B

This is a Qwen3-VL-Thinking 8B vision-language reasoning model trained with the VIG method. It is intended for efficient visual reasoning and reproducible research.

Model Details

  • Architecture: Qwen3VLForConditionalGeneration
  • Transformers model type: qwen3_vl
  • Precision: bfloat16
  • Input: image and text
  • Output: generated text
  • Weights: complete Safetensors checkpoint

Usage

import torch
from transformers import Qwen3VLForConditionalGeneration, AutoProcessor

model_id = "chaser2/VIG-Qwen3-VL-Thinking-8B"
model = Qwen3VLForConditionalGeneration.from_pretrained(
    model_id, torch_dtype=torch.bfloat16, device_map="auto"
)
processor = AutoProcessor.from_pretrained(model_id)

messages = [{
    "role": "user",
    "content": [
        {"type": "image", "image": "path/to/image.jpg"},
        {"type": "text", "text": "Describe the image and explain your reasoning."}
    ]
}]
prompt = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[prompt], images=["path/to/image.jpg"], return_tensors="pt", padding=True).to(model.device)
with torch.inference_mode():
    outputs = model.generate(**inputs, max_new_tokens=1024)
answer = outputs[:, inputs.input_ids.shape[1]:]
print(processor.batch_decode(answer, skip_special_tokens=True)[0])

The repository contains the complete model weights, configuration, tokenizer, processor, and generation configuration for reproducible use.

Downloads last month
73
Safetensors
Model size
9B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for chaser2/VIG-Qwen3-VL-Thinking-8B

Quantizations
1 model