Improve language tag
#3
by
lbourdois
- opened
README.md
CHANGED
|
@@ -1,150 +1,161 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
language:
|
| 4 |
-
-
|
| 5 |
-
-
|
| 6 |
-
|
| 7 |
-
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
-
|
| 11 |
-
-
|
| 12 |
-
-
|
| 13 |
-
|
| 14 |
-
-
|
| 15 |
-
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
|
| 50 |
-
|
|
| 51 |
-
|
|
| 52 |
-
|
|
| 53 |
-
|
|
| 54 |
-
|
|
| 55 |
-
|
|
| 56 |
-
|
|
| 57 |
-
|
|
| 58 |
-
|
|
| 59 |
-
|
|
| 60 |
-
|
|
| 61 |
-
|
|
| 62 |
-
|
|
| 63 |
-
|
|
| 64 |
-
|
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 150 |
```
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- zho
|
| 5 |
+
- eng
|
| 6 |
+
- fra
|
| 7 |
+
- spa
|
| 8 |
+
- por
|
| 9 |
+
- deu
|
| 10 |
+
- ita
|
| 11 |
+
- rus
|
| 12 |
+
- jpn
|
| 13 |
+
- kor
|
| 14 |
+
- vie
|
| 15 |
+
- tha
|
| 16 |
+
- ara
|
| 17 |
+
tags:
|
| 18 |
+
- multimodal
|
| 19 |
+
library_name: transformers
|
| 20 |
+
datasets:
|
| 21 |
+
- BAAI/Infinity-MM
|
| 22 |
+
- BAAI/Infinity-Instruct
|
| 23 |
+
- BAAI/Infinity-Preference
|
| 24 |
+
base_model:
|
| 25 |
+
- Qwen/Qwen2.5-1.5B-Instruct
|
| 26 |
+
- google/siglip-so400m-patch14-384
|
| 27 |
+
pipeline_tag: visual-question-answering
|
| 28 |
+
---
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
# Introduction
|
| 33 |
+
|
| 34 |
+
The **Aquila-VL-2B** model is a vision-language model (VLM) trained based on the [LLava-one-vision](https://llava-vl.github.io/blog/2024-08-05-llava-onevision/) framework. The [Qwen2.5-1.5B-instruct](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct) model is chose as the LLM, while [siglip-so400m-patch14-384](https://huggingface.co/google/siglip-so400m-patch14-384) is utilized as the vision tower.
|
| 35 |
+
|
| 36 |
+
The model was trained on our self-built Infinity-MM dataset, which contains approximately 40 million image-text pairs. This dataset is a combination of open-source data collected from the internet and synthetic instruction data generated using open-source VLM models.
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
We have open-sourced [Infinity-MM](https://huggingface.co/datasets/BAAI/Infinity-MM) dataset and related resources. We hope you enjoy using them!
|
| 40 |
+
|
| 41 |
+
## News
|
| 42 |
+
- `2024/11/19`: We have released [intermediate checkpoints](https://huggingface.co/BAAI/Aquila-VL-2B-Intermediate) obtained during different stages of training. Please feel free to use these models for analysis and experimentation.
|
| 43 |
+
- `2024/10/25`: The [Aquila-VL-2B](https://huggingface.co/BAAI/Aquila-VL-2B-llava-qwen) model and [Infinity-MM](https://huggingface.co/datasets/BAAI/Infinity-MM) dataset are now available. We have also released the [technical report](https://arxiv.org/abs/2410.18558) simultaneously.
|
| 44 |
+
|
| 45 |
+
# Evaluation
|
| 46 |
+
|
| 47 |
+
We evaluated the model using the [VLMEvalKit](https://github.com/open-compass/VLMEvalKit) tool. Whenever possible, we prioritized using the OpenAI API for test sets that support API-based evaluation.
|
| 48 |
+
|
| 49 |
+
| Benchmark | MiniCPM-V-2 | InternVL2-2B | XinYuan-VL-2B | Qwen2-VL-2B-Instruct | Aquila-VL-2B |
|
| 50 |
+
| :--------------------------- | :---------: | :----------: | :-----------: | :------------------: | :----------: |
|
| 51 |
+
| MMBench-EN<sub>test</sub> | 69.4 | 73.4 | **78.9** | 74.9 | 78.8 |
|
| 52 |
+
| MMBench-CN<sub>test</sub> | 65.9 | 70.9 | 76.1 | 73.9 | **76.4** |
|
| 53 |
+
| MMBench_V1.1<sub>test</sub> | 65.2 | 69.7 | **75.4** | 72.7 | 75.2 |
|
| 54 |
+
| MMT-Bench<sub>test</sub> | 54.5 | 53.3 | 57.2 | 54.8 | **58.2** |
|
| 55 |
+
| RealWorldQA | 55.4 | 57.3 | 63.9 | 62.6 | **63.9** |
|
| 56 |
+
| HallusionBench | 36.8 | 38.1 | 36.0 | 41.5 | **43.0** |
|
| 57 |
+
| SEEDBench2<sub>plus</sub> | 51.8 | 60.0 | 63.0 | 62.4 | **63.0** |
|
| 58 |
+
| LLaVABench | 66.1 | 64.8 | 42.4 | 52.5 | **68.4** |
|
| 59 |
+
| MMStar | 41.6 | 50.2 | 51.9 | 47.8 | **54.9** |
|
| 60 |
+
| POPE | 86.6 | 85.3 | **89.4** | 88.0 | 83.6 |
|
| 61 |
+
| MMVet | 44.0 | 41.1 | 42.7 | **50.7** | 44.3 |
|
| 62 |
+
| MMMU<sub>val</sub> | 39.6 | 34.9 | 43.6 | 41.7 | **47.4** |
|
| 63 |
+
| ScienceQA<sub>test</sub> | 80.4 | 94.1 | 86.6 | 78.1 | **95.2** |
|
| 64 |
+
| AI2D<sub>test</sub> | 64.8 | 74.4 | 74.2 | 74.6 | **75.0** |
|
| 65 |
+
| MathVista<sub>testmini</sub> | 39.0 | 45.0 | 47.1 | 47.9 | **59.0** |
|
| 66 |
+
| MathVerse<sub>testmini</sub> | 19.8 | 24.7 | 22.2 | 21.0 | **26.2** |
|
| 67 |
+
| MathVision | 15.4 | 12.6 | 16.3 | 17.5 | **18.4** |
|
| 68 |
+
| DocVQA<sub>test</sub> | 71.0 | 86.9 | 87.6 | **89.9** | 85.0 |
|
| 69 |
+
| InfoVQA<sub>test</sub> | 40.0 | 59.5 | 59.1 | **65.4** | 58.3 |
|
| 70 |
+
| ChartQA<sub>test</sub> | 59.6 | 71.4 | 57.1 | 73.5 | **76.5** |
|
| 71 |
+
| TextVQA<sub>val</sub> | 74.3 | 73.5 | 77.6 | **79.9** | 76.4 |
|
| 72 |
+
| OCRVQA<sub>testcore</sub> | 54.4 | 40.2 | 67.6 | **68.7** | 64.0 |
|
| 73 |
+
| VCR<sub>en easy</sub> | 27.6 | 51.6 | 67.7 | 68.3 | **70.0** |
|
| 74 |
+
| OCRBench | 613 | 784 | 782 | **810** | 772 |
|
| 75 |
+
| Average | 53.5 | 58.8 | 60.9 | 62.1 | **64.1** |
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
|
| 79 |
+
For comparison models, evaluations were conducted in a local environment, so the scores may differ slightly from those reported in papers or on the official VLMEvalKit leaderboard.
|
| 80 |
+
|
| 81 |
+
# How to use
|
| 82 |
+
|
| 83 |
+
```python
|
| 84 |
+
# pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git
|
| 85 |
+
from llava.model.builder import load_pretrained_model
|
| 86 |
+
from llava.mm_utils import process_images, tokenizer_image_token
|
| 87 |
+
from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN
|
| 88 |
+
from llava.conversation import conv_templates
|
| 89 |
+
from PIL import Image
|
| 90 |
+
import requests
|
| 91 |
+
import copy
|
| 92 |
+
import torch
|
| 93 |
+
import warnings
|
| 94 |
+
|
| 95 |
+
warnings.filterwarnings("ignore")
|
| 96 |
+
|
| 97 |
+
pretrained = "BAAI/Aquila-VL-2B-llava-qwen"
|
| 98 |
+
|
| 99 |
+
model_name = "llava_qwen"
|
| 100 |
+
device = "cuda"
|
| 101 |
+
device_map = "auto"
|
| 102 |
+
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
|
| 103 |
+
|
| 104 |
+
model.eval()
|
| 105 |
+
|
| 106 |
+
# load image from url
|
| 107 |
+
url = "https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true"
|
| 108 |
+
image = Image.open(requests.get(url, stream=True).raw)
|
| 109 |
+
|
| 110 |
+
# load image from local environment
|
| 111 |
+
# url = "./local_image.jpg"
|
| 112 |
+
# image = Image.open(url)
|
| 113 |
+
|
| 114 |
+
image_tensor = process_images([image], image_processor, model.config)
|
| 115 |
+
image_tensor = [_image.to(dtype=torch.float16, device=device) for _image in image_tensor]
|
| 116 |
+
|
| 117 |
+
conv_template = "qwen_1_5" # Make sure you use correct chat template for different models
|
| 118 |
+
question = DEFAULT_IMAGE_TOKEN + "\nWhat is shown in this image?"
|
| 119 |
+
conv = copy.deepcopy(conv_templates[conv_template])
|
| 120 |
+
conv.append_message(conv.roles[0], question)
|
| 121 |
+
conv.append_message(conv.roles[1], None)
|
| 122 |
+
prompt_question = conv.get_prompt()
|
| 123 |
+
|
| 124 |
+
input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(device)
|
| 125 |
+
image_sizes = [image.size]
|
| 126 |
+
|
| 127 |
+
cont = model.generate(
|
| 128 |
+
input_ids,
|
| 129 |
+
images=image_tensor,
|
| 130 |
+
image_sizes=image_sizes,
|
| 131 |
+
do_sample=False,
|
| 132 |
+
temperature=0,
|
| 133 |
+
max_new_tokens=4096,
|
| 134 |
+
)
|
| 135 |
+
|
| 136 |
+
text_outputs = tokenizer.batch_decode(cont, skip_special_tokens=True)
|
| 137 |
+
|
| 138 |
+
print(text_outputs)
|
| 139 |
+
```
|
| 140 |
+
|
| 141 |
+
|
| 142 |
+
|
| 143 |
+
# Future Plan
|
| 144 |
+
|
| 145 |
+
* We plan to train models of various sizes.
|
| 146 |
+
* Future training will incorporate multi-image and video data.
|
| 147 |
+
|
| 148 |
+
|
| 149 |
+
## **Citation**
|
| 150 |
+
If you find this useful, please cite the following work
|
| 151 |
+
```
|
| 152 |
+
@misc{gu2024infinitymmscalingmultimodalperformance,
|
| 153 |
+
title={Infinity-MM: Scaling Multimodal Performance with Large-Scale and High-Quality Instruction Data},
|
| 154 |
+
author={Shuhao Gu and Jialing Zhang and Siyuan Zhou and Kevin Yu and Zhaohu Xing and Liangdong Wang and Zhou Cao and Jintao Jia and Zhuoyi Zhang and Yixuan Wang and Zhenchong Hu and Bo-Wen Zhang and Jijie Li and Dong Liang and Yingli Zhao and Yulong Ao and Yaoqi Liu and Fangxiang Feng and Guang Liu},
|
| 155 |
+
year={2024},
|
| 156 |
+
eprint={2410.18558},
|
| 157 |
+
archivePrefix={arXiv},
|
| 158 |
+
primaryClass={cs.CL},
|
| 159 |
+
url={https://arxiv.org/abs/2410.18558},
|
| 160 |
+
}
|
| 161 |
```
|