Learning multi-granularities visual tokens a coarse-to-fine nested way
Mu Cai, Jianwei Yang, Jianfeng Gao , Yong Jae Lee
[Paper] [Project Page] [Demo] [Model Zoo]
- [6/3] 🔥 All training (
llava-1.5-m3) and evaluations (llava-1.5-m3andllava-next-m3) code are release. - [5/27] 🔥 We released Matryoshka Multimodal Models. We propose to learn visual tokens in a nested manner following a coarse-to-fine order. Checkout the paper and demo.
The fundamental implementation of M3 can be found in this code snippet.
If you are not using Linux, do NOT proceed, see instructions for macOS and Windows.
- Clone this repository and navigate to LLaVA folder
git clone https://github.com/mu-cai/matryoshka-mm.git
cd matryoshka-mm- Install Package
conda create -n matryoshka-mm python=3.10 -y
conda activate matryoshka-mm
pip install --upgrade pip # enable PEP 660 support
pip install -e .- Install additional packages for training cases
pip install -e ".[train]"
pip install flash-attn --no-build-isolation
Example Code
from llava.model.builder import load_pretrained_model
from llava.mm_utils import get_model_name_from_path
from llava.eval.run_llava import eval_model
model_path = "mucai/llava-next-vicuna-7b-m3"
tokenizer, model, image_processor, context_len = load_pretrained_model(
model_path=model_path,
model_base=None,
model_name=get_model_name_from_path(model_path)
)Check out the details wth the load_pretrained_model function in llava/model/builder.py.
You can also use the eval_model function in llava/eval/run_llava.py to get the output easily. By doing so, you can use this code on Colab directly after downloading this repository.
model_path = "mucai/llava-next-vicuna-7b-m3"
prompt = "What are the things I should be cautious about when I visit here?"
image_file = "https://llava-vl.github.io/static/images/view.jpg"
args = type('Args', (), {
"model_path": model_path,
"model_base": None,
"model_name": get_model_name_from_path(model_path),
"query": prompt,
"conv_mode": None,
"image_file": image_file,
"sep": ",",
"temperature": 0,
"top_p": None,
"num_beams": 1,
"max_new_tokens": 512,
"matryoshka_vis_token_scale": 576,
})()
eval_model(args)Please check out our Model Zoo for all public M3 checkpoints, and the instructions of how to use the weights.
To launch a Gradio demo locally, please run the following commands one by one. If you plan to launch multiple model workers to compare between different checkpoints, you only need to launch the controller and the web server ONCE.
flowchart BT
%% Declare Nodes
gws("Gradio (UI Server)")
c("Controller (API Server):<br/>PORT: 10000")
mw7b("Model Worker:<br/>llava-next-vicuna-7b-m3<br/>PORT: 40000")
mw13b("Model Worker:<br/>llava-next-vicuna-7b-m3<br/>PORT: 40001")
sglw13b("Backend:<br/>llava-v1.5-7b-m3<br/>http://localhost:30000")
lsglw13b("Worker:<br/>lllava-v1.5-7b-m3<<br/>PORT: 40002")
%% Declare Styles
classDef data fill:#3af,stroke:#48a,stroke-width:2px,color:#444
classDef success fill:#8f8,stroke:#0a0,stroke-width:2px,color:#444
classDef failure fill:#f88,stroke:#f00,stroke-width:2px,color:#444
%% Assign Styles
class id,od data;
class cimg,cs_s,scsim_s success;
class ncimg,cs_f,scsim_f failure;
subgraph Demo Connections
direction BT
c<-->gws
mw7b<-->c
mw13b<-->c
lsglw13b<-->c
sglw13b<-->lsglw13b
end
python -m llava.serve.controller --host 0.0.0.0 --port 30000python -m llava.serve.gradio_web_server --controller http://localhost:30000 --model-list-mode reloadYou just launched the Gradio web interface. 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.
This is the actual worker that performs the inference on the GPU. Each worker is responsible for a single model specified in --model-path.
python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:30000 --port 40000 --worker http://localhost:40000 --model-path mucai/llava-next-vicuna-7b-m3Wait until the process finishes loading the model and you see "Uvicorn running on ...". Now, refresh your Gradio web UI, and you will see the model you just launched in the model list.
You can launch as many workers as you want, and compare between different model checkpoints in the same Gradio interface. Please keep the --controller the same, and modify the --port and --worker to a different port number for each worker.
python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:30000 --port <different from 40000, say 40001> --worker http://localhost:<change accordingly, i.e. 40001> --model-path <ckpt2>If you are using an Apple device with an M1 or M2 chip, you can specify the mps device by using the --device flag: --device mps.
If the VRAM of your GPU is less than 24GB (e.g., RTX 3090, RTX 4090, etc.), you may try running it with multiple GPUs. Our latest code base will automatically try to use multiple GPUs if you have more than one GPU. You can specify which GPUs to use with CUDA_VISIBLE_DEVICES. Below is an example of running with the first two GPUs.
CUDA_VISIBLE_DEVICES=0,1 python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:30000 --port 40000 --worker http://localhost:40000 --model-path mucai/llava-next-vicuna-7b-m3You can launch the model worker with quantized bits (4-bit, 8-bit), which allows you to run the inference with reduced GPU memory footprint, potentially allowing you to run on a GPU with as few as 12GB VRAM. Note that inference with quantized bits may not be as accurate as the full-precision model. Simply append --load-4bit or --load-8bit to the model worker command that you are executing. Below is an example of running with 4-bit quantization.
python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:30000 --port 40000 --worker http://localhost:40000 --model-path mucai/llava-next-vicuna-7b-m3 --load-4bitYou can train and launch the model worker with LoRA weights using our instructions here..
Chat about images using LLaVA without the need of Gradio interface. It also supports multiple GPUs, 4-bit and 8-bit quantized inference. With 4-bit quantization, for our LLaVA-1.5-7B, it uses less than 8GB VRAM on a single GPU.
python -m llava.serve.cli \
--model-path mucai/llava-next-vicuna-7b-m3 \
--image-file "https://llava-vl.github.io/static/images/view.jpg" \
--matryoshka_vis_token_scale 576 \
--load-4bit
M3 finetunes LLaVA checkpoints using the exact same visual instruction data.
LLaVA is trained on 8 H100 GPUs with 80GB memory. 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: per_device_train_batch_size x gradient_accumulation_steps x num_gpus.
We use the exact same hyperparameters as LLaVA in finetuning. Hyperparameters used are provided below.
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|---|---|---|---|---|---|
| LLaVA-v1.5-7B-M3 | 128 | 2e-5 | 1 | 2048 | 0 |
Our base model Vicuna v1.5, which is an instruction-tuned chatbot, will be downloaded automatically when you run our provided training scripts. No action is needed.
- Prepare data
Please download the annotation of the final mixture our instruction tuning data llava_v1_5_mix665k.json, and download the images from constituting datasets:
- COCO: train2017
- GQA: images
- OCR-VQA: download script, we save all files as
.jpg - TextVQA: train_val_images
- VisualGenome: part1, part2
After downloading all of them, organize the data as follows in ./playground/data,
├── coco
│ └── train2017
├── gqa
│ └── images
├── ocr_vqa
│ └── images
├── textvqa
│ └── train_images
└── vg
├── VG_100K
└── VG_100K_2
- Start training!
You may download our pretrained projectors in Model Zoo. It is not recommended to use legacy projectors, as they may be trained with a different version of the codebase, and if any option is off, the model will not function/train as we expected.
Training script with DeepSpeed ZeRO-3: finetune.sh.
If you are do not have enough GPU memory:
- Use LoRA:
finetune_lora.sh. We are able to fit 13B training in 8-A100-40G/8-A6000, and 7B training in 8-RTX3090. Make sureper_device_train_batch_size*gradient_accumulation_stepsis the same as the provided script for best reproducibility. - Replace
zero3.jsonwithzero3_offload.jsonwhich offloads some parameters to CPU RAM. This slows down the training speed.
If you are interested in finetuning M3 model to your own task/data, please check out Finetune_Custom_Data.md。
We use the same benchmark as LLaVA-1.5 and LLaVA-Next
For LLaVA-1.5, see Evaluation.md.
For LLaVA-NeXT on image understanding, see lmms-eval.
For LLaVA-NeXT on video understanding, see IG-VLM.
If you find LLaVA useful for your research and applications, please cite using this BibTeX:
@article{cai2024matryoshka,
title={Matryoshka Multimodal Models},
author={Cai, Mu and Yang, Jianwei and Gao, Jianfeng and Lee, Yong Jae},
journal={arXiv preprint arXiv:2405.17430},
year={2024}
}