SliMM: A Simple LMM baseline with Dynamic Visual Resolution 🚀

Embracing Dynamic Visual Resolution with LLaVA-style training!

[🌐 Project Page] [📚 Paper] [🤗 Checkpoints] [🤖 demo]

🌟 Highlights

• Advanced Techniques: We incorporate native dynamic resolution, as used in Qwen2-VL, for high-resolution visual encoding, replacing the previous cumbersome Multi-Crop/AnyRes methods. Moreover, building on DeepStack [1], we maintain the same principle of interting stacked visual tokens into multiple layers of the LLMs. We propose two enhanced versions for native resolution vision encoding: DeepStack-MidLayers, which improves performance with negligible additional FLOPs by stacking multi-level visual tokens from the middle layers of the vision encoder, and DeepStack-Efficient, which reduces visual token usage while maintaining high performance.

• Seamless Integration: Easily use LLaVA-format training data in our codebase.

• Training Efficiency: Fine-tuning on the 748K LLaVA-Next-DATA for on epoch takes only 4 hours for 0.5/2B Qwen2 and 6 hours for a 7B on 8xH100, which is more than 2x faster than LLaVA-OV codebase.

• Strong Baseline Model for Small LMMs: We establish a robust baseline using widely-used public available datasets, including LCS-758K (Stage-1), LLaVA-OV-MidStage (Stage 1.5), and LLaVA-OneVision SI (Stage 2).

  [1] DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs

🔥 News

• [2024/12/12] Our first version is out! We release a strong 0.5B baseline model SliMM-Qwen2-0.5B and advanced baseline SliMM-DeepStackM-Qwen2-0.5B. We release a strong 2B model SliMM-DeepStackE-Qwen2VL-2B continous fine-tuned from Qwen2VL-2B, which save 4x fewer visual tokens for LLM with. Training scrips are avaliable here!

🛠️ Installation

1. Clone this repository and navigate to SliMM folder

git clone https://github.com/MengLcool/SliMM.git
cd SliMM

2. Install Package

conda create -n slimm python=3.10 -y
conda activate slimm
pip install --upgrade pip  # enable PEP 660 support
pip install -e .

# It is currently crucial to install the transformers library with commit 7bbc62474391aff64f63fcc064c975752d1fa4de
# This issue will be resolved in the next version.
pip install transformers@git+https://github.com/huggingface/transformers.git@7bbc62474391aff64f63fcc064c975752d1fa4de

# additional packages for training cases
pip install -e ".[train]"
pip install flash-attn --no-build-isolation

Quick Start With HuggingFace

Example Code

# this is very similar to qwen2-vl
from slimm.model.processor import SliMMQwen2VLProcessor
from slimm.model.slimm import SliMMForConditionalGeneration
from slimm.model.utils_vl import process_vision_info

model_path = "menglc/SliMM-DeepStackE-Qwen2VL-2B"

model = SliMMForConditionalGeneration.from_pretrained(
    model_path, torch_dtype="auto", device_map="auto"
)

processor = SliMMQwen2VLProcessor.from_pretrained(model_path)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Describe this image."},
        ],
    }
]

# Preparation for inference
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")

# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)

📝 Todo

• Release training code and checkpoints.
• Instruction tuning on videos and multiple images.
• Improve training efficiency with adavanced multimodal packing.
• Distillation from large LMM to small LMM.

🔗 Citation

If you find our work helpful, please consider citing our paper 📎 and starring our repo 🌟 :

@inproceedings{meng2024deepstack,
  title={DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs},
  author={Meng, Lingchen and Yang, Jianwei and Tian, Rui and Dai, Xiyang and Wu, Zuxuan and Gao, Jianfeng and Jiang, Yu-Gang},
  booktitle={NeurIPS},
  year={2024}
}

Acknowledgement

Our work is built upon Qwen2-VL, LLaVA and LLaVA-NeXT.

✨ Feel free to contribute and reach out if you have any questions!