Instructions to use HughieHu/DR-Pixel-Trained with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use HughieHu/DR-Pixel-Trained with Transformers:

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

pipe = pipeline("image-text-to-text", model="HughieHu/DR-Pixel-Trained")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("HughieHu/DR-Pixel-Trained")
model = AutoModelForMultimodalLM.from_pretrained("HughieHu/DR-Pixel-Trained", device_map="auto")
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?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use HughieHu/DR-Pixel-Trained with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "HughieHu/DR-Pixel-Trained"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "HughieHu/DR-Pixel-Trained",
		"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/HughieHu/DR-Pixel-Trained

SGLang

How to use HughieHu/DR-Pixel-Trained 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 "HughieHu/DR-Pixel-Trained" \
    --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": "HughieHu/DR-Pixel-Trained",
		"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 "HughieHu/DR-Pixel-Trained" \
        --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": "HughieHu/DR-Pixel-Trained",
		"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"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use HughieHu/DR-Pixel-Trained with Docker Model Runner:
```
docker model run hf.co/HughieHu/DR-Pixel-Trained
```

DR Pixel Trained

This is the trained DR Pixel model from DR Pixel: Page-Native Deep Research with Vision-Language Models: Qwen3.6-27B fine-tuned to answer questions about scientific papers by searching over and reading their rendered pages. It was trained with counterfactual credit reweighting on the 2,496 GPT-5.4 trajectories in the train/ directory of HughieHu/drpixel-bench, and evaluated on the benchmark in the same repository.

Files

model-00001-of-00012.safetensors …   the fine-tuned weights (bf16), the model evaluated in the paper
adapter/                              the same fine-tune as a LoRA adapter on Qwen/Qwen3.6-27B

The root holds the base model with the LoRA adapter merged in, so it serves like any Qwen3.6-27B checkpoint. adapter/ holds the adapter alone (rank 32, alpha 64, all linear layers of the language model), for loading on top of the base model with PEFT or for further training.

Evaluating on DRPixelBench

Serve the model with any OpenAI-compatible server (vLLM, SGLang, …) with a 131,072-token context, and run the harness from github.com/HughieHu/DRPixel (eval/) against it. The paper's runs used the sampling settings below; the harness passes them through when they are set.

MODEL_NAME=HughieHu/DR-Pixel-Trained
MODEL_BASE_URL=http://localhost:8000/v1
MODEL_API_KEY=EMPTY
MODEL_TEMPERATURE=0.6 MODEL_TOP_P=0.95 MODEL_TOP_K=20 MODEL_PRESENCE_PENALTY=1.0
SEARCH_URL=https://drpixel--drpixel-search-gate.us-east.modal.run    # hosted page search

python -m drbench.eval --config Page

Training

LoRA on all linear layers of the language model, one epoch at learning rate 1e-4 with an effective batch size of 16 and a maximum length of 131,072 tokens, with per-step loss weights from counterfactual credit and a constant weight of 1.5 on the final answer. train/README.md describes the method and the full recipe.

License

The weights are released under Apache 2.0, the license of the base model. The training data are trajectories generated by GPT-5.4 and remain subject to the terms of that model.

Citation

@misc{hu2026drpixel,
  title  = {DR Pixel: Page-Native Deep Research with Vision-Language Models},
  author = {Hu, Tiansheng and Wang, Yichuan and Talwar, James V. and Tomlin, Nicholas and Zhao, Chen and Min, Sewon},
  year   = {2026}
}

Downloads last month: 23

Safetensors

Model size

27B params

Tensor type

BF16

Model tree for HughieHu/DR-Pixel-Trained

Base model

Qwen/Qwen3.6-27B

Finetuned

(421)

this model

HughieHu
/

DR-Pixel-Trained