🌋 LLaVA: Large Language and Vision Assistant

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

[📢LLaVA-NeXT Blog] [Project Page] [Demo] [Data] [Model Zoo]

🤝Community Contributions: [llama.cpp] [Colab] [🤗Space] [Replicate] [AutoGen] [BakLLaVA]

Improved Baselines with Visual Instruction Tuning[Paper] [HF]
Haotian Liu,Chunyuan Li,Yuheng Li,Yong Jae Lee

Visual Instruction Tuning(NeurIPS 2023,Oral) [Paper] [HF]
Haotian Liu*,Chunyuan Li*,Qingyang Wu,Yong Jae Lee(*Equal Contribution)

Release

[2024/05/10] 🔥LLaVA-NeXT(Stronger) models are released, stronger LMM with support of LLama-3 (8B) and Qwen-1.5 (72B/110B). [Blog] [Checkpoints] [Demo] [Code]
[2024/05/10] 🔥LLaVA-NeXT(Video) is released. The image-only-trained LLaVA-NeXT model is surprisingly strong on video tasks with zero-shot modality transfer. DPO training with AI feedback on videos can yield significant improvement. [Blog] [Checkpoints] [Code]
[03/10] ReleasingLMMs-Eval,a highly efficient evaluation pipeline we used when developing LLaVA-NeXT. It supports the evaluation of LMMs on dozens of public datasets and allows new dataset onboarding, making the dev of new LMMs much faster. [Blog] [Codebase]
[1/30] 🔥LLaVA-NeXT(LLaVA-1.6) is out! With additional scaling to LLaVA-1.5, LLaVA-NeXT-34B outperforms Gemini Pro on some benchmarks. It can now process 4x more pixels and perform more tasks/applications than before. Check out theblog post,and explore thedemo!Models are available inModel Zoo.Training/eval data and scripts coming soon.
[11/10]LLaVA-Plusis released: Learning to Use Tools for Creating Multimodal Agents, with LLaVA-Plus (LLaVA that Plug and Learn to Use Skills). [Project Page] [Demo] [Code] [Paper]
[11/2]LLaVA-Interactiveis released: Experience the future of human-AI multimodal interaction with an all-in-one demo for Image Chat, Segmentation, Generation and Editing. [Project Page] [Demo] [Code] [Paper]
[10/26] 🔥 LLaVA-1.5 with LoRA achieves comparable performance as full-model finetuning, with a reduced GPU RAM requirement (ckpts,script). We also provide adocon how to finetune LLaVA-1.5 on your own dataset with LoRA.
[10/12] Check out the Korean LLaVA (Ko-LLaVA), created by ETRI, who has generously supported our research! [🤗 Demo]
[10/5] 🔥 LLaVA-1.5 is out! Achieving SoTA on 11 benchmarks, with just simple modifications to the original LLaVA, utilizes all public data, completes training in ~1 day on a single 8-A100 node, and surpasses methods like Qwen-VL-Chat that use billion-scale data. Check out thetechnical report,and explore thedemo!Models are available inModel Zoo.The training data and scripts of LLaVA-1.5 are releasedhere,and evaluation scripts are releasedhere!
[9/26] LLaVA is improved with reinforcement learning from human feedback (RLHF) to improve fact grounding and reduce hallucination. Check out the new SFT and RLHF checkpoints at project[LLavA-RLHF]
[9/22]LLaVAis accepted by NeurIPS 2023 asoral presentation,andLLaVA-Medis accepted by NeurIPS 2023 Datasets and Benchmarks Track asspotlight presentation.

More

[11/6] SupportInteldGPU and CPU platforms.More details here.
[10/12] LLaVA is now supported inllama.cppwith 4-bit / 5-bit quantization support!
[10/11] The training data and scripts of LLaVA-1.5 are releasedhere,and evaluation scripts are releasedhere!
[10/10]Roboflow Deep Dive:First Impressions with LLaVA-1.5.
[9/20] We summarize our empirical study of training 33B and 65B LLaVA models in anote.Further, if you are interested in the comprehensive review, evolution and trend of multimodal foundation models, please check out our recent survey paper``Multimodal Foundation Models: From Specialists to General-Purpose Assistants''.

[7/19] 🔥 We release a major upgrade, including support for LLaMA-2, LoRA training, 4-/8-bit inference, higher resolution (336x336), and a lot more. We releaseLLaVA Benchfor benchmarking open-ended visual chat with results from Bard and Bing-Chat. We also support and verify training with RTX 3090 and RTX A6000. Check outLLaVA-from-LLaMA-2,and ourmodel zoo!
[6/26]CVPR 2023 TutorialonLarge Multimodal Models: Towards Building and Surpassing Multimodal GPT-4!Please check out [Slides] [Notes] [YouTube] [Bilibli].
[6/11] We released the preview for the most requested feature: DeepSpeed and LoRA support! Please see documentationshere.
[6/1] We releasedLLaVA-Med: Large Language and Vision Assistant for Biomedicine,a step towards building biomedical domain large language and vision models with GPT-4 level capabilities. Checkout thepaperandpage.
[5/6] We are releasingLLaVA-Lighting-MPT-7B-preview,based on MPT-7B-Chat! Seeherefor more details.
[5/2] 🔥 We are releasing LLaVA-Lighting! Train a lite, multimodal GPT-4 with just $40 in 3 hours! Seeherefor more details.
[4/27] Thanks to the community effort, LLaVA-13B with 4-bit quantization allows you to run on a GPU with as few as 12GB VRAM! Try it outhere.
[4/17] 🔥 We releasedLLaVA: Large Language and Vision Assistant.We propose visual instruction tuning, towards building large language and vision models with GPT-4 level capabilities. Checkout thepaperanddemo.

Usage and License Notices: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, including but not limited to theOpenAI Terms of Usefor the dataset and the specific licenses for base language models for checkpoints trained using the dataset (e.g.Llama community licensefor LLaMA-2 and Vicuna-v1.5). This project does not impose any additional constraints beyond those stipulated in the original licenses. Furthermore, users are reminded to ensure that their use of the dataset and checkpoints is in compliance with all applicable laws and regulations.

Install

If you are not using Linux, doNOTproceed, see instructions formacOSandWindows.

Clone this repository and navigate to LLaVA folder

git clone https://github /haotian-liu/LLaVA.git
cdLLaVA

Install Package

conda create -n llava Python =3.10 -y
conda activate llava
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

Upgrade to latest code base

git pull
pip install -e.

#if you see some import errors when you upgrade,
#please try running the command below (without #)
#pip install flash-attn --no-build-isolation --no-cache-dir

Quick Start With HuggingFace

Example Code

fromllava.model.builderimportload_pretrained_model
fromllava.mm_utilsimportget_model_name_from_path
fromllava.eval.run_llavaimporteval_model

model_path="liuhaotian/llava-v1.5-7b"

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 theload_pretrained_modelfunction inllava/model/builder.py.

You can also use theeval_modelfunction inllava/eval/run_llava.pyto get the output easily. By doing so, you can use this code on Colab directly after downloading this repository.

model_path="liuhaotian/llava-v1.5-7b"
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
})()

eval_model(args)

LLaVA Weights

Please check out ourModel Zoofor all public LLaVA checkpoints, and the instructions of how to use the weights.

Demo

Gradio Web UI

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 serverONCE.

flowchart BT
%% Declare Nodes
gws( "Gradio (UI Server)" )
c( "Controller (API Server):<br/>PORT: 10000" )
mw7b( "Model Worker:<br/>llava-v1.5-7b<br/>PORT: 40000" )
mw13b( "Model Worker:<br/>llava-v1.5-13b<br/>PORT: 40001" )
sglw13b( "SGLang Backend:<br/>llava-v1.6-34b<br/>http://localhost:30000" )
lsglw13b( "SGLang Worker:<br/>llava-v1.6-34b<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

Launch a controller

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

Launch a gradio web server.

Python -m llava.serve.gradio_web_server --controller http://localhost:10000 --model-list-mode reload

You 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.

Launch a SGLang worker

This is the recommended way to serve LLaVA model with high throughput, and you need to install SGLang first. Note that currently4-bitquantization is not supported yet on SGLang-LLaVA, and if you have limited GPU VRAM, please check out model worker withquantization.

pip install"sglang[all]"

You'll first launch a SGLang backend worker which will execute the models on GPUs. Remember the--portyou've set and you'll use that later.

#Single GPU
CUDA_VISIBLE_DEVICES=0 Python 3 -m sglang.launch_server --model-path liuhaotian/llava-v1.5-7b --tokenizer-path llava-hf/llava-1.5-7b-hf --port 30000

#Multiple GPUs with tensor parallel
CUDA_VISIBLE_DEVICES=0,1 Python 3 -m sglang.launch_server --model-path liuhaotian/llava-v1.5-13b --tokenizer-path llava-hf/llava-1.5-13b-hf --port 30000 --tp 2

Tokenizers (temporary):llava-hf/llava-1.5-7b-hf,llava-hf/llava-1.5-13b-hf,liuhaotian/llava-v1.6-34b-tokenizer.

You'll then launch a LLaVA-SGLang worker that will communicate between LLaVA controller and SGLang backend to route the requests. Set--sgl-endpointtohttp://127.0.0.1:portwhereportis the one you just set (default: 30000).

Python -m llava.serve.sglang_worker --host 0.0.0.0 --controller http://localhost:10000 --port 40000 --worker http://localhost:40000 --sgl-endpoint http://127.0.0.1:30000

Launch a model worker

This is the actualworkerthat 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:10000 --port 40000 --worker http://localhost:40000 --model-path liuhaotian/llava-v1.5-13b

Wait 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--controllerthe same, and modify the--portand--workerto a different port number for each worker.

Python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:10000 --port<different from 40000, say40001>--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--deviceflag:--device mps.

Launch a model worker (Multiple GPUs, when GPU VRAM <= 24GB)

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 withCUDA_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:10000 --port 40000 --worker http://localhost:40000 --model-path liuhaotian/llava-v1.5-13b

Launch a model worker (4-bit, 8-bit inference, quantized)

You 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-4bitor--load-8bitto themodel workercommand 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:10000 --port 40000 --worker http://localhost:40000 --model-path liuhaotian/llava-v1.5-13b --load-4bit

Launch a model worker (LoRA weights, unmerged)

You can launch the model worker with LoRA weights, without merging them with the base checkpoint, to save disk space. There will be additional loading time, while the inference speed is the same as the merged checkpoints. Unmerged LoRA checkpoints do not havelora-mergein the model name, and are usually much smaller (less than 1GB) than the merged checkpoints (13G for 7B, and 25G for 13B).

To load unmerged LoRA weights, you simply need to pass an additional argument--model-base,which is the base LLM that is used to train the LoRA weights. You can check the base LLM of each LoRA weights in themodel zoo.

Python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:10000 --port 40000 --worker http://localhost:40000 --model-path liuhaotian/llava-v1-0719-336px-lora-vicuna-13b-v1.3 --model-base lmsys/vicuna-13b-v1.3

CLI Inference

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 liuhaotian/llava-v1.5-7b \
--image-file"https://llava-vl.github.io/static/images/view.jpg"\
--load-4bit

Train

Below is the latest training configuration for LLaVA v1.5. For legacy models, please refer to README ofthisversion for now. We'll add them in a separate doc later.

LLaVA training consists of two stages: (1) feature alignment stage: use our 558K subset of the LAION-CC-SBU dataset to connect afrozen pretrainedvision encoder to afrozen LLM;(2) visual instruction tuning stage: use 150K GPT-generated multimodal instruction-following data, plus around 515K VQA data from academic-oriented tasks, to teach the model to follow multimodal instructions.

LLaVA is trained on 8 A100 GPUs with 80GB memory. To train on fewer GPUs, you can reduce theper_device_train_batch_sizeand increase thegradient_accumulation_stepsaccordingly. Always keep the global batch size the same:per_device_train_batch_sizexgradient_accumulation_stepsxnum_gpus.

Hyperparameters

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-v1.5-13B	256	1e-3	1	2048	0

Finetuning

Hyperparameter	Global Batch Size	Learning rate	Epochs	Max length	Weight decay
LLaVA-v1.5-13B	128	2e-5	1	2048	0

Download Vicuna checkpoints (automatically)

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.

Pretrain (feature alignment)

Please download the 558K subset of the LAION-CC-SBU dataset with BLIP captions we use in the paperhere.

Pretrain takes around 5.5 hours for LLaVA-v1.5-13B on 8x A100 (80G), due to the increased resolution to 336px. It takes around 3.5 hours for LLaVA-v1.5-7B.

Training script with DeepSpeed ZeRO-2:pretrain.sh.

--mm_projector_type mlp2x_gelu:the two-layer MLP vision-language connector.
--vision_tower openai/clip-vit-large-patch14-336:CLIP ViT-L/14 336px.

Pretrain takes around 20 hours for LLaVA-7B on 8x V100 (32G)

We provide training script with DeepSpeedhere. Tips:

If you are using V100 which is not supported by FlashAttention, you can use thememory-efficient attentionimplemented inxFormers.Install xformers and replacellava/train/train_mem.pyabove withllava/train/train_xformers.py.

Visual Instruction Tuning

Prepare data

Please download the annotation of the final mixture our instruction tuning datallava_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 inModel 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.

Visual instruction tuning takes around 20 hours for LLaVA-v1.5-13B on 8x A100 (80G), due to the increased resolution to 336px. It takes around 10 hours for LLaVA-v1.5-7B on 8x A100 (40G).

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.
Replacezero3.jsonwithzero3_offload.jsonwhich offloads some parameters to CPU RAM. This slows down the training speed.

If you are interested in finetuning LLaVA model to your own task/data, please check outFinetune_Custom_Data.md.

New options to note:

--mm_projector_type mlp2x_gelu:the two-layer MLP vision-language connector.
--vision_tower openai/clip-vit-large-patch14-336:CLIP ViT-L/14 336px.
--image_aspect_ratio pad:this pads the non-square images to square, instead of cropping them; it slightly reduces hallucination.
--group_by_modality_length True:this should only be used when your instruction tuning dataset contains both language (e.g. ShareGPT) and multimodal (e.g. LLaVA-Instruct). It makes the training sampler only sample a single modality (either image or language) during training, which we observe to speed up training by ~25%, and does not affect the final outcome.

Evaluation

In LLaVA-1.5, we evaluate models on a diverse set of 12 benchmarks. To ensure the reproducibility, we evaluate the models with greedy decoding. We do not evaluate using beam search to make the inference process consistent with the chat demo of real-time outputs.

SeeEvaluation.md.

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-path./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-our.jsonl

Evaluate the generated responses. In our case,answer-file-ref.jsonlis the response generated by text-only GPT-4 (0314), with the context captions/boxes provided.

OPENAI_API_KEY="sk-***********************************"Python llava/eval/eval_gpt_review_visual.py \
--question playground/data/coco2014_val_qa_eval/qa90_questions.jsonl \
--context llava/eval/table/caps_boxes_coco2014_val_80.jsonl \
--answer-list \
/path/to/answer-file-ref.jsonl \
/path/to/answer-file-our.jsonl \
--rule llava/eval/table/rule.json \
--output /path/to/review.json

Summarize the evaluation results

Python summarize_gpt_review.py

Citation

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

@misc{liu2024llavanext,
title={LLaVA-NeXT: Improved reasoning, OCR, and world knowledge},
url={https://llava-vl.github.io/blog/2024-01-30-llava-next/},
author={Liu, Haotian and Li, Chunyuan and Li, Yuheng and Li, Bo and Zhang, Yuanhan and Shen, Sheng and Lee, Yong Jae},
month={January},
year={2024}
}

@misc{liu2023improvedllava,
title={Improved Baselines with Visual Instruction Tuning},
author={Liu, Haotian and Li, Chunyuan and Li, Yuheng and Lee, Yong Jae},
publisher={arXiv:2310.03744},
year={2023},
}

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

Acknowledgement

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

Related Projects

For future project ideas, please check out:

SEEM: Segment Everything Everywhere All at Once
Grounded-Segment-Anythingto detect, segment, and generate anything by marryingGrounding DINOandSegment-Anything.

Name		Name	Last commit message	Last commit date
Latest commit History 460 Commits
.devcontainer		.devcontainer
.github/ISSUE_TEMPLATE		.github/ISSUE_TEMPLATE
docs		docs
images		images
llava		llava
playground/data		playground/data
scripts		scripts
.dockerignore		.dockerignore
.editorconfig		.editorconfig
.gitattributes		.gitattributes
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
cog.yaml		cog.yaml
predict.py		predict.py
pyproject.toml		pyproject.toml

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

🌋 LLaVA: Large Language and Vision Assistant

Release

Contents

Install

Upgrade to latest code base

Quick Start With HuggingFace

LLaVA Weights

Demo

Gradio Web UI

Launch a controller

Launch a gradio web server.

Launch a SGLang worker

Launch a model worker

Launch a model worker (Multiple GPUs, when GPU VRAM <= 24GB)

Launch a model worker (4-bit, 8-bit inference, quantized)

Launch a model worker (LoRA weights, unmerged)

CLI Inference

Train

Hyperparameters

Download Vicuna checkpoints (automatically)

Pretrain (feature alignment)

Visual Instruction Tuning

Evaluation

GPT-assisted Evaluation

Citation

Acknowledgement

Related Projects

About

Releases 6

Contributors 48

Languages

License

haotian-liu/LLaVA

Folders and files

Latest commit

History

Repository files navigation

🌋 LLaVA: Large Language and Vision Assistant

Release

Contents

Install

Upgrade to latest code base

Quick Start With HuggingFace

LLaVA Weights

Demo

Gradio Web UI

Launch a controller

Launch a gradio web server.

Launch a SGLang worker

Launch a model worker

Launch a model worker (Multiple GPUs, when GPU VRAM <= 24GB)

Launch a model worker (4-bit, 8-bit inference, quantized)

Launch a model worker (LoRA weights, unmerged)

CLI Inference

Train

Hyperparameters

Download Vicuna checkpoints (automatically)

Pretrain (feature alignment)

Visual Instruction Tuning

Evaluation

GPT-assisted Evaluation

Citation

Acknowledgement

Related Projects

About

Topics

Resources

License

Stars

Watchers

Forks

Releases 6

Contributors 48

Languages