🌋 LLaVA: Large Language and Vision Assitant

Visual instruction tuning towards large language and vision models with GPT-4 level capabilities.

[Project Page] [Paper] [Demo] [Data] [Model]

Visual Instruction Tuning
Haotian Liu*, Chunyuan Li*, Qingyang Wu, Yong Jae Lee (*Equal Contribution)

Release

🔥 We released LLaVA: Large Language and Vision Assitant. We propose visual instruction tuning, towards building large language and vision models with GPT-4 level capabilities. Checkout the paper and demo.

Usage and License Notices: The data, code and checkpoint is intended and licensed for research use only. They are also restricted to uses that follow the license agreement of LLaMA, Vicuna and GPT-4. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes.

Data Donwnload
Install
LLaVA Weights
Serving
Evaluation
Fine-tuning

Data Donwnload

Data file name	Size
conversation_58k.json	126 MB
detail_23k.json	20.5 MB
complex_reasoning_77k.json	79.6 MB

To download our langauge-image multimodal instruction-folllowing dataset LLaVA-Instruct-150K, please run the following script:

sh download_data.sh

Install

Clone this repository and navigate to LLaVA folder

git clone https://github.com/haotian-liu/LLaVA.git
cd LLaVA

Install Package

pip install --upgrade pip  # enable PEP 660 support
pip install -e .

# For training
pip install ninja
pip install flash-attn

NOTE: In this research preview, we used a modified version of huggingface/transformers library to support multimodal models and the LLaMA tokenizer. Make sure that you are using the correct transformers library from https://github.com/haotian-liu/transformers_llava.

LLaVA Weights

We release LLaVA weights as delta weights to comply with the LLaMA model license. You can add our delta to the original LLaMA weights to obtain the LLaVA weights. Instructions:

Get the original LLaMA weights in the huggingface format by following the instructions here.
Use the following scripts to get LLaVA weights by applying our delta. It will automatically download delta weights from our Hugging Face account.

LLaVA-13B

This conversion command needs around 60 GB of CPU RAM.

python3 -m llava.model.apply_delta \
    --base /path/to/llama-13b \
    --target /output/path/to/LLaVA-13B-v0 \
    --delta liuhaotian/LLaVA-13b-delta-v0

LLaVA-7B

Coming soon.

Serving

Web UI

Launch a controller

python -m llava.serve.controller --host 0.0.0.0 --port 10000

Launch a model worker

python -m llava.serve.model_worker --controller http://localhost:10000 --port 40000 --worker http://localhost:40000 --model-path ./checkpoints/LLaVA-13B-v0 --multi-modal

Wait until the process finishes loading the model and you see "Uvicorn running on ...".

Send a test message

python -m llava.serve.test_message --model-name LLaVA-13B-v0 --controller http://localhost:10000

Launch a gradio web server.

python -m llava.serve.gradio_web_server --controller http://localhost:10000

You can open your browser and chat with a model now.

Evaluation

GPT-assisted Evaluation

Our GPT-assisted evaluation pipeline for multimodal modeling is provided for a comprehensive understanding of the capabilities of vision-language models. Please see our paper for more details.

Generate LLaVA responses

python model_vqa.py \
    --model-name ./checkpoints/LLaVA-13B-v0 \
    --question-file \
    playground/data/coco2014_val_qa_eval/qa90_questions.jsonl \
    --image-folder \
    /path/to/coco2014_val \
    --answers-file \
    /path/to/answer-file.jsonl

Evaluate the generated responses

OPENAI_API_KEY="sk-***********************************" python eval_gpt_review_visual.py \
    --question playground/data/coco2014_val_qa_eval/qa90_questions.jsonl \
    --context table/caps_boxes_coco2014_val_80.jsonl \
    --answer-list \
    /path/to/answer-file-1.jsonl \
    /path/to/answer-file-2.jsonl \
    --rule table/rule.json \
    --output /path/to/review.json

Summarize the evaluation results

python summarize_gpt_review.py

ScienceQA

Please see ScienceQA repo for setting up the dataset. You may either use the official ScienceQA evaluation script or our script to evaluate the model.

Fine-tuning

Data

The current version of LLaVA is fine-tuned from a Vicuna-13B model. We use approximately 600K filtered CC3M in feature alignment pretraining and 150K GPT-generated multimodal instruction-following data in finetuning. For detailed description of the data generation pipeline, please refer see our paper.

We are working on a more capable model that is pretrained with the data at a larger scale. Stay tuned!

We release all three types of multimodal instruction-following data. The use of these data is subject to OpenAI TOS.

Code and Hyperparameters

We fine-tune the model using the code from FastChat. We use a similar set of hyperparameters as Vicuna in finetuning. Both hyperparameters used in pretraining and finetuning are provided below.

Pretraining

Hyperparameter	Global Batch Size	Learning rate	Epochs	Max length	Weight decay
LLaVA-13B	128	2e-3	1	2048	0

Finetuning

Hyperparameter	Global Batch Size	Learning rate	Epochs	Max length	Weight decay
LLaVA-13B	32	2e-5	3	2048	0

Fine-tuning with Local GPUs

LLaVA is trained on 8 A100 GPUs with 80GB memory with the following code. To train on fewer GPUs, you can reduce the per_device_train_batch_size and increase the gradient_accumulation_steps accordingly to keep the global batch size the same.

Pretraining

torchrun --nnodes=1 --nproc_per_node=8 --master_port=25001 \
    fastchat/train/train_mem.py \
    --model_name_or_path ./checkpoints/llama-vicuna-13b \
    --data_path /path/to/cc3m_595k.json \
    --image_folder /path/to/cc3m_595k \
    --vision_tower openai/clip-vit-large-patch14 \
    --tune_mm_mlp_adapter True \
    --mm_vision_select_layer -2 \
    --mm_use_im_start_end \
    --bf16 True \
    --output_dir ./checkpoints/llava-13b-pretrain \
    --num_train_epochs 1 \
    --per_device_train_batch_size 16 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 1 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 2400 \
    --save_total_limit 1 \
    --learning_rate 2e-3 \
    --weight_decay 0. \
    --warmup_ratio 0.03 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --tf32 True \
    --model_max_length 2048 \
    --gradient_checkpointing True \
    --lazy_preprocess True \
    --report_to wandb

Extract projector features

python scripts/extract_mm_projector.py \
  --model_name_or_path ./checkpoints/llava-13b-pretrain \
  --output ./checkpoints/mm_projector/llava-13b-pretrain.bin

Finetuning

torchrun --nnodes=1 --nproc_per_node=8 --master_port=25001 \
    fastchat/train/train_mem.py \
    --model_name_or_path /path/to/llama-vicuna-13b \
    --data_path /path/to/llava_instruct_150k.json \
    --image_folder /Data/haotian/coco/train2017 \
    --vision_tower openai/clip-vit-large-patch14 \
    --pretrain_mm_mlp_adapter ./checkpoints/mm_projector/llava-13b-pretrain.bin \
    --mm_vision_select_layer -2 \
    --mm_use_im_start_end True \
    --bf16 True \
    --output_dir ./checkpoints \
    --num_train_epochs 3 \
    --per_device_train_batch_size 4 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 1 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 5000 \
    --save_total_limit 3 \
    --learning_rate 2e-5 \
    --weight_decay 0. \
    --warmup_ratio 0.03 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --tf32 True \
    --fsdp "full_shard auto_wrap" \
    --fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
    --model_max_length 2048 \
    --gradient_checkpointing True \
    --lazy_preprocess True \
    --report_to wandb

Acknowledgement

Vicuna: the codebase we built upon, and our base model Vicuna-13B that has the amazing language capabilities!

If you find LLaVA useful for your your research and applications, please cite using this BibTeX:

@misc{liu2023llava,
      title={Visual Instruction Tuning}, 
      author={Liu, Haotian and Li, Chunyuan and Wu, Qingyang and Lee, Yong Jae},
      publisher={arXiv:2304.08485},
      year={2023},
}

Related Projects

For future project ideas, checkout the Grounded-Segment-Anything to detect, segment, and generate anything by marrying Grounding DINO and Segment-Anything.

TobyGE/LLaVA