|
--- |
|
datasets: |
|
- lmms-lab/LLaVA-OneVision-Data |
|
language: |
|
- en |
|
- zh |
|
library_name: transformers |
|
license: apache-2.0 |
|
metrics: |
|
- accuracy |
|
tags: |
|
- multimodal |
|
--- |
|
|
|
# LLaVA-OneVision |
|
|
|
 |
|
|
|
Play with the model on the [LLaVA OneVision Chat](https://llava-onevision.lmms-lab.com/). |
|
|
|
## Table of Contents |
|
|
|
1. [Model Summary](##model-summary) |
|
2. [Use](##use) |
|
3. [Limitations](##limitations) |
|
4. [Training](##training) |
|
5. [License](##license) |
|
6. [Citation](##citation) |
|
|
|
## Model Summary |
|
|
|
`llava-onevision-72b-ov-chat` is our latest model specifically designed for chat scenarios. It is built upon `llava-onevision-72b-ov` and has undergone iterative DPO training with human preference, making it well-suited for chat applications. |
|
|
|
Research by [Tianyi Xiong](https://tyxiong23.github.io/) indicates that our iterative DPO training method enhances the model's chat capabilities while preserving its instruction-following abilities. |
|
|
|
For further details, please refer to our upcoming blog or paper. |
|
|
|
- **Repository:** [LLaVA-VL/LLaVA-NeXT](https://github.com/LLaVA-VL/LLaVA-NeXT?tab=readme-ov-file) |
|
- **Project Website:** [llava-onevision.lmms-lab.com](llava-onevision.lmms-lab.com) |
|
- **Paper:** [LLaVA-OneVision](arxiv.org/abs/2408.03326) |
|
- **Point of Contact:** [Tianyi Xiong](https://tyxiong23.github.io/), [Bo Li](mailto:drluodian@gmail.com) |
|
- **Languages:** English, Chinese |
|
|
|
## Benchmark Performance |
|
|
|
To be released |
|
|
|
## Use |
|
|
|
### Intended use |
|
|
|
The model was trained on [LLaVA-OneVision Dataset](https://huggingface.co/datasets/lmms-lab/LLaVA-OneVision-Data) and have the ability to interact with images, multi-image and videos. |
|
|
|
**Feel free to share your generations in the Community tab!** |
|
|
|
### Generation |
|
|
|
```python |
|
# pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git |
|
from llava.model.builder import load_pretrained_model |
|
from llava.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token |
|
from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, IGNORE_INDEX |
|
from llava.conversation import conv_templates, SeparatorStyle |
|
|
|
from PIL import Image |
|
import requests |
|
import copy |
|
import torch |
|
|
|
import sys |
|
import warnings |
|
|
|
warnings.filterwarnings("ignore") |
|
pretrained = "lmms-lab/llava-onevision-qwen2-0.5b-si" |
|
model_name = "llava_qwen" |
|
device = "cuda" |
|
device_map = "auto" |
|
tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, device_map=device_map) # Add any other thing you want to pass in llava_model_args |
|
|
|
model.eval() |
|
|
|
url = "https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true" |
|
image = Image.open(requests.get(url, stream=True).raw) |
|
image_tensor = process_images([image], image_processor, model.config) |
|
image_tensor = [_image.to(dtype=torch.float16, device=device) for _image in image_tensor] |
|
|
|
conv_template = "qwen_1_5" # Make sure you use correct chat template for different models |
|
question = DEFAULT_IMAGE_TOKEN + "\nWhat is shown in this image?" |
|
conv = copy.deepcopy(conv_templates[conv_template]) |
|
conv.append_message(conv.roles[0], question) |
|
conv.append_message(conv.roles[1], None) |
|
prompt_question = conv.get_prompt() |
|
|
|
input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(device) |
|
image_sizes = [image.size] |
|
|
|
|
|
cont = model.generate( |
|
input_ids, |
|
images=image_tensor, |
|
image_sizes=image_sizes, |
|
do_sample=False, |
|
temperature=0, |
|
max_new_tokens=4096, |
|
) |
|
text_outputs = tokenizer.batch_decode(cont, skip_special_tokens=True) |
|
print(text_outputs) |
|
``` |
|
|
|
# Training |
|
|
|
## Model |
|
|
|
- **Architecture:** SO400M + Qwen2 |
|
- **Pretraining Stage:** LCS-558K, 1 epoch, projector |
|
- **Mid Stage:** A mixture of 4.7M high-quality synthetic data, 1 epoch, full model |
|
- **Final-Image Stage:** A mixture of 3.6M single-image data, 1 epoch, full model |
|
- **OneVision Stage:** A mixture of 1.6M single-image/multi-image/video data, 1 epoch, full model |
|
- **Precision:** bfloat16 |
|
|
|
## Hardware & Software |
|
|
|
- **GPUs:** 256 \* Nvidia Tesla A100 (for whole model series training) |
|
- **Orchestration:** [Huggingface Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) |
|
- **Neural networks:** [PyTorch](https://github.com/pytorch/pytorch) |
|
|
|
# Citation |
|
|
|
``` |
|
@article{li2024llavaonevision, |
|
title={LLaVA-OneVision}, |
|
} |
|
``` |
|
|
