Bunny: A family of lightweight multimodal models

📖 Technical report | 🤗 Bunny-v1.0-3B | 🤖 ModelScope | 🧠 WiseModel | 🤗 Data | 🤖 Data | 🐰 Demo

Bunny is a family of lightweight but powerful multimodal models. It offers multiple plug-and-play vision encoders, like EVA-CLIP, SigLIP and language backbones, including Phi-1.5, StableLM-2 and Phi-2. To compensate for the decrease in model size, we construct more informative training data by curated selection from a broader data source. Remarkably, our Bunny-v1.0-3B model built upon SigLIP and Phi-2 outperforms the state-of-the-art MLLMs, not only in comparison with models of similar size but also against larger MLLMs (7B), and even achieves performance on par with 13B models.

News and Updates

• 2024.03.06 🔥 Bunny training data is released! Check more details about Bunny-v1.0-data in HuggingFace or ModelScope!
• 2024.02.20 🔥 Bunny technical report is ready! Check more details about Bunny here!
• 2024.02.07 🔥 Bunny is released! Bunny-v1.0-3B built upon SigLIP and Phi-2 outperforms the state-of-the-art MLLMs, not only in comparison with models of similar size but also against larger MLLMs (7B), and even achieves performance on par with LLaVA-13B!

Quickstart

HuggingFace transformers

Here we show a code snippet to show you how to use Bunny-v1.0-3B with HuggingFace transformers:

import torch
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image
import warnings

# disable some warnings
transformers.logging.set_verbosity_error()
transformers.logging.disable_progress_bar()
warnings.filterwarnings('ignore')

# set device
torch.set_default_device('cpu')  # or 'cuda'

# create model
model = AutoModelForCausalLM.from_pretrained(
    'BAAI/Bunny-v1_0-3B',
    torch_dtype=torch.float16,
    device_map='auto',
    trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(
    'BAAI/Bunny-v1_0-3B',
    trust_remote_code=True)

# text prompt
prompt = 'Why is the image funny?'
text = f"A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: <image>\n{prompt} ASSISTANT:"
text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0)

# image, sample images can be found in https://huggingface.co/BAAI/Bunny-v1_0-3B/tree/main/images
image = Image.open('example_2.png')
image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)

# generate
output_ids = model.generate(
    input_ids,
    images=image_tensor,
    max_new_tokens=100,
    use_cache=True)[0]

print(tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip())

Before running the snippet, you need to install the following dependencies:

pip install torch transformers accelerate pillow

ModelScope

We advise users especially those in Chinese mainland to use ModelScope. snapshot_download can help you solve issues concerning downloading checkpoints.

Expand to see the snippet

import torch
import transformers
from modelscope import AutoTokenizer, AutoModelForCausalLM
from modelscope.hub.snapshot_download import snapshot_download
from PIL import Image
import warnings

# disable some warnings
transformers.logging.set_verbosity_error()
transformers.logging.disable_progress_bar()
warnings.filterwarnings('ignore')

# set device
torch.set_default_device('cpu')  # or 'cuda'

# create model
snapshot_download(model_id='thomas/siglip-so400m-patch14-384')
model = AutoModelForCausalLM.from_pretrained(
    'BAAI/Bunny-v1.0-3B',
    torch_dtype=torch.float16,
    device_map='auto',
    trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(
    'BAAI/Bunny-v1.0-3B',
    trust_remote_code=True)

# text prompt
prompt = 'Why is the image funny?'
text = f"A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: <image>\n{prompt} ASSISTANT:"
text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0)

# image, sample images can be found in images folder on https://www.modelscope.cn/models/BAAI/Bunny-v1.0-3B/files
image = Image.open('example_2.png')
image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)

# generate
output_ids = model.generate(
    input_ids,
    images=image_tensor,
    max_new_tokens=100,
    use_cache=True)[0]

print(tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip())

Before running the snippet, you need to install the following dependencies:

pip install torch modelscope transformers accelerate pillow

Model Zoo

• Evaluation

Checkpoint	MME$^\text{P}$	MME$^\text{C}$	MMB$^\text{T}$	MMB$^\text{D}$	SEED	MMMU$^\text{V}$	MMMU$^\text{T}$	VQA$^\text{v2}$	GQA	SQA$^\text{I}$	POPE
bunny-phi-1.5-eva-lora	1213.7	278.9	60.9	56.8	56.4	30.0	28.4	76.5	60.4	58.2	86.1
bunny-stablelm-2-eva-lora	1301.0	235.0	58.4	56.4	55.3	29.8	29.4	74.6	56.7	60.0	84.8
bunny-phi-2-eva-lora	1421.0	285.4	68.6	67.4	62.2	35.9	32.6	78.9	62.3	69.1	87.1
bunny-phi-1.5-siglip-lora	1230.0	237.5	61.2	59.7	57.7	30.0	29.1	78.0	61.1	61.3	85.8
bunny-stablelm-2-siglip-lora	1366.8	236.1	65.1	62.8	58.8	29.9	29.8	78.9	60.9	61.1	85.9
Bunny-v1.0-3B/bunny-phi-2-siglip	1488.8	289.3	69.2	68.6	62.5	38.2	33.0	79.8	62.5	70.9	86.8

The model with the best performance is denoted as Bunny-v1.0-3B or bunny-phi-2-siglip, whose merged weights can be found here and the LoRA weights can be found here.

• Training details

Checkpoint	Vision Encoder	LLM	Pretrain lr	Pretrain weights
bunny-phi-1.5-eva-lora	EVA02_CLIP_L_336_psz14_s6B	microsoft/phi-1_5	1e-3	bunny-pretrain-phi-1.5-eva
bunny-stablelm-2-eva-lora	EVA02_CLIP_L_336_psz14_s6B	stabilityai/stablelm-2-1_6b	1e-3	bunny-pretrain-stablelm-2-eva
bunny-phi-2-eva-lora	EVA02_CLIP_L_336_psz14_s6B	microsoft/phi-2	5e-5	bunny-pretrain-phi-2-eva
bunny-phi-1.5-siglip-lora	siglip-so400m-patch14-384	microsoft/phi-1_5	5e-4	bunny-pretrain-phi-1.5-siglip
bunny-stablelm-2-siglip-lora	siglip-so400m-patch14-384	stabilityai/stablelm-2-1_6b	5e-4	bunny-pretrain-stablelm-2-siglip
bunny-phi-2-siglip-lora	siglip-so400m-patch14-384	microsoft/phi-2	5e-4	bunny-pretrain-phi-2-siglip

Install

• CUDA and cuDNN

  We use CUDA 11.8 and cuDNN 8.7.0. We actually use the CUDA docker by NVIDIA: docker pull nvcr.io/nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04. CUDA 12 is fine, too.

• Create a conda virtual environment and activate it:

  conda create -n bunny python=3.10
  conda activate bunny

• Basic requirements

  pip install --upgrade pip  # enable PEP 660 support
  pip install transformers
  pip install torch torchvision xformers --index-url https://download.pytorch.org/whl/cu118

• Install apex

  # https://github.com/NVIDIA/apex#from-source
  pip install ninja
  git clone https://github.com/NVIDIA/apex
  cd apex
  # if pip >= 23.1 (ref: https://pip.pypa.io/en/stable/news/#v23-1) which supports multiple `--config-settings` with the same key...
  pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./
  # otherwise
  pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --global-option="--cpp_ext" --global-option="--cuda_ext" ./

• Install flash-attention

  # https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#installation-and-features
  pip install packaging
  pip install flash-attn --no-build-isolation

• Install bunny and other requirements

  git clone https://github.com/BAAI-DCAI/Bunny.git
  cd Bunny
  pip install -e .

Training

Bunny training consists of two stages: (1) pretrain stage: use data to connect a frozen pretrained vision encoder to a frozen LLM, and only the connector is trained; (2) visual instruction tuning stage: use data to teach the model to follow multimodal instructions, where the connector and learnable LLM parameters are updated.

Bunny is trained on 8 A100 GPUs. To train on fewer GPUs, you can reduce the per_device_train_batch_size and increase the gradient_accumulation_steps accordingly. Always keep the global batch size the same: global_batch_size = per_device_train_batch_size $\times$ gradient_accumulation_steps $\times$ num_gpus.

Support Models

Currently, we support several vision encoders and LLMs.

For vision encoders, we support CLIP, EVA-CLIP and SigLIP.

Vision Encoders	Download Link
clip-vit-large-patch14-336	openai/clip-vit-large-patch14-336
EVA02_CLIP_L_336_psz14_s6B	QuanSun/EVA-CLIP
siglip-so400m-patch14-384	google/siglip-so400m-patch14-384

For LLMs, we support phi-1.5, stablelm-2 and phi-2.

MODEL_TYPE	LLM	Download Link
phi-1.5	phi-1_5	microsoft/phi-1_5
stablelm-2	stablelm-2-1_6b	stabilityai/stablelm-2-1_6b
phi-2	phi-2	microsoft/phi-2

Note that there are many variants of above models. We build and test our code based on the exact versions mentioned above. More models will be supported in the future!

Pretrain

• Data preparation

  We use a high-quality coreset with less duplicates and more informative samples of LAION-2B built by this work. We randomly sample 2 million image-text pairs from the coreset and convert them to training format. The dataset is available here.

• Run

  Update --model_name_or_path and --vision_tower to the paths of the LLM and vision encoder, respectively. Update MODEL_TYPE and OUTPUT_DIR accordingly. The global batch size is 256. The optimal learning rate varies for different settings and we list the lr in our experiments in the Model Zoo.

  sh script/train/pretrain.sh

Visual Instruction Tuning

• Data preparation

  We build Bunny-695K by modifying SVIT-mix-665K for finetuning. The dataset is available here.

• Run

  Update --model_name_or_path and --vision_tower to the paths of the LLM and vision encoder, respectively. Update MODEL_TYPE, PRETRAIN_DIR and OUTPUT_DIR accordingly. The global batch size is 128.

  # full-parameter tuning
  sh script/train/finetune_full.sh
  
  # LoRA tuning
  sh script/train/finetune_lora.sh

Demo

Gradio Web UI

• Starting the Controller

  First, start the controller. This service orchestrates communication between the web server and model workers.

  python -m bunny.serve.controller \
  	--host 0.0.0.0 \
  	--port 10000

• Launching the Gradio Web Server

  To interact with the models through a web interface, start the Gradio web server.

  Basic start:

  python -m bunny.serve.gradio_web_server \
  	--controller http://localhost:10000 \
  	--model-list-mode reload

  If you want to share your web server with others, use --share option. Note that frpc_linux_amd64_v0.2 may be missing and you can fix it following instructions printed on the screen.

  python -m bunny.serve.gradio_web_server \
  	--controller http://localhost:10000 \
  	--model-list-mode reload \
  	--share

  Now, you can open the web interface with the URL printed on the screen. You may notice that there is no model in the model list. Do not worry, as we have not launched any model worker yet. It will be automatically updated when you launch a model worker.

• Launching Model Workers

  Model workers handle the processing of model inferences. Configure each worker with the appropriate model and start it.

  • For full-parameter tuning models

    python -m bunny.serve.model_worker \
      --host 0.0.0.0 \
      --controller http://localhost:10000 \
      --port 40000 \
      --worker http://localhost:40000 \
      --model-path /path/to/bunny/model \
      --model-type phi-2 (or stablelm-2 or phi-1.5)

  • For LoRA tuning models

    You can use script/merge_lora_weights.py to merge the LoRA weights and base LLM, and use it as above.

    python script/merge_lora_weights.py \
      --model-path /path/to/bunny_lora_weights \
      --model-base /path/to/base_llm_model \
      --model-type phi-2 (or stablelm-2 or phi-1.5) \
      --save-model-path /path/to/merged_model

    Or you can use it without merging as below.

    python -m bunny.serve.model_worker \
      --host 0.0.0.0 \
      --controller http://localhost:10000 \
      --port 40000 \
      --worker http://localhost:40000 \
      --model-path /path/to/bunny_lora_weights \
      --model-base /path/to/base_llm_model \
      --model-type phi-2 (or stablelm-2 or phi-1.5)

CLI Inference (Without Gradio Interface)

For CLI-based inference without using the Gradio interface, use the following command:

• For full-parameter tuning models

  python -m bunny.serve.cli \
  	--model-path /path/to/bunny/model \
  	--model-type phi-2 (or stablelm-2 or phi-1.5) \
  	--image-file /path/to/the/test/image

• For LoRA tuning models

  You can use script/merge_lora_weights.py to merge the LoRA weights and base LLM, and use it as above.

  python script/merge_lora_weights.py \
  	--model-path /path/to/bunny_lora_weights \
  	--model-base /path/to/base_llm_model \
  	--model-type phi-2 (or stablelm-2 or phi-1.5) \
  	--save-model-path /path/to/merged_model

  Or you can use it without merging as below.

  python -m bunny.serve.cli \
  	--model-path /path/to/bunny_lora_weights \
  	--model-base /path/to/base_llm_model \
  	--model-type phi-2 (or stablelm-2 or phi-1.5) \
  	--image-file /path/to/the/test/image

Evaluation

For full-parameter tuning models, see evaluation_full.md.

For LoRA tuning models, see evaluation_lora.md.

Citation

If you find this repository helpful, please cite the paper below.

@article{he2024bunny,
      title={Efficient Multimodal Learning from Data-centric Perspective}, 
      author={He, Muyang and Liu, Yexin and Wu, Boya and Yuan, Jianhao and Wang, Yueze and Huang, Tiejun and Zhao, Bo},
      journal={arXiv preprint arXiv:2402.11530},
      year={2024}
}

License

This project utilizes certain datasets and checkpoints that are subject to their respective original licenses. Users must comply with all terms and conditions of these original licenses. The content of this project itself is licensed under the Apache license 2.0.

Acknowledgement

We build our project based on LLaVA: Large Language and Vision Assistant.