Instructions to use Infi-MM/infimm-hd with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Infi-MM/infimm-hd with Transformers:

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

pipe = pipeline("text-generation", model="Infi-MM/infimm-hd", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Infi-MM/infimm-hd", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use Infi-MM/infimm-hd with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Infi-MM/infimm-hd"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Infi-MM/infimm-hd",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Infi-MM/infimm-hd

SGLang

How to use Infi-MM/infimm-hd 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 "Infi-MM/infimm-hd" \
    --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": "Infi-MM/infimm-hd",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "Infi-MM/infimm-hd" \
        --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": "Infi-MM/infimm-hd",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Infi-MM/infimm-hd with Docker Model Runner:
```
docker model run hf.co/Infi-MM/infimm-hd
```

Paper

More detailes can be found in our paper at https://arxiv.org/abs/2403.01487. We have released the pretraining model and the pyotrch code at https://github.com/InfiMM/infimm-hd/. Feel free to build your model from our pretrained model.

Quickstart

Use the code below to get started with the base model:

import torch
from transformers import AutoModelForCausalLM, AutoProcessor

processor = AutoProcessor.from_pretrained("Infi-MM/infimm-hd", trust_remote_code=True)

prompts = [
    {
        "role": "user",
        "content": [
            {"image": "/xxx/test.jpg"}, # change it with you image
            "Please describe the image in detail.",
        ],
    }
]
inputs = processor(prompts)
# use bf16 and gpu 0
model = AutoModelForCausalLM.from_pretrained(
    "Infi-MM/infimm-hd",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
).to(0).eval()

inputs = inputs

inputs["batch_images"] = inputs["batch_images"].to(torch.bfloat16)
for k in inputs:
    inputs[k] = inputs[k].to(model.device)

generated_ids = model.generate(
    **inputs,
    min_new_tokens=0,
    max_new_tokens=256,
)
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)
print(generated_text)

License

This project is licensed under the CC BY-NC 4.0.

The copyright of the images belongs to the original authors.

See LICENSE for more information.

Contact Us

Please feel free to contact us via email [email protected] if you have any questions.

Citation

@misc{liu2024infimmhd,
      title={InfiMM-HD: A Leap Forward in High-Resolution Multimodal Understanding}, 
      author={Haogeng Liu and Quanzeng You and Xiaotian Han and Yiqi Wang and Bohan Zhai and Yongfei Liu and Yunzhe Tao and Huaibo Huang and Ran He and Hongxia Yang},
      year={2024},
      eprint={2403.01487},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Downloads last month: 701

Datasets used to train Infi-MM/infimm-hd

Paper for Infi-MM/infimm-hd

InfiMM-HD: A Leap Forward in High-Resolution Multimodal Understanding

Paper • 2403.01487 • Published Mar 3, 2024 • 16