MLM_Filter

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

★ 71

Official implementation of our paper "Finetuned Multimodal Language Models are High-Quality Image-Text Data Filters".

Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# MLM Filter

Official implementation of our paper "[Finetuned Multimodal Language Models are High-Quality Image-Text Data Filters](https://arxiv.org/pdf/2403.02677.pdf)". <br>

## Release
- [12/30/2024] 🔥 We released a new generation MLM-Filter model based on Qwen2.5-1.5B, [mlm-filter-qwen2.5-1.5b-gpt4o](https://huggingface.co/weizhiwang/mlm-filter-qwen2.5-1.5b-gpt4o). The instruction data are re-generated with GPT-4o. With the much smaller LLM backbone, the inference has been significantly improved. The llava codebase for mlm-filter model inference has been completely removed and integrated into [LLaVA-Unified](https://github.com/Victorwz/LLaVA-Unified).
- [10/24/2024] 🔥 We released two new MLM-Filter models based on llama3, [mlm-filter-llama-3-8b](https://huggingface.co/weizhiwang/mlm-filter-llama-3-8b) and [mlm-filter-llama-3.2-3b](https://huggingface.co/weizhiwang/mlm-filter-llama-3.2-3b).
- [2/25/2024] 🔥 We released **Finetuned Multimodal Language Models are High-Quality Image-Text Data Filters**. We propose to adopt fine-tuned Multimodal Language Model as effective and efficient data filters to select high-quality image-text pairs from large-scale web-crawled iamge-text data. Checkout the [paper](https://arxiv.org/pdf/2403.02677.pdf).

## Project Structure

- [mlm_filter_scoring_single_image.py](mlm_filter_scoring_single_image.py): Sample code for perform quality score generation on a single image-text pair
- [mlm_filter_scoring_datacomp_batch_inference.py](mlm_filter_scoring_datacomp_batch_inference.py): Sample code for perform large-scale quality score generation on Webdataset format image-text data
- [mlm_filter_scoring_datacomp_batch_inference_v2.py](mlm_filter_scoring_datacomp_batch_inference_v2.py): Sample code for perform large-scale quality score generation on Webdataset format image-text data for Llama3 or Qwen2.5 based MLM-Filter models
- [run_inference.sh](run_inference.sh): Sample code for perform large-scale quality score generation on Webdataset format image-text data on machines with 8 GPUs

## Install

We highly suggest you to use python==3.10, i.e.,
```bash
conda create -n mlm_filter python=3.10
```
Then install the dependencies for quality score generation:
```bash
pip install git+https://github.com/Victorwz/LLaVA-Unified.git
```

<!-- ### 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.

```Shell
python 
``` -->

## Quality Score Generation

### Inference on Single Image

```Shell
python mlm_filter_scoring_single_image.py --image-path /path/to/image --caption "text caption"
```
Parameters to note:

- `--metric`: quality scoring metric for generation, select among `image_text_matching`, `object_detail_fulfillment`, `caption_text_quality`, `semantic_understanding`, `all`
- `--image-path`: path to image file or image url
- `--caption`: text caption

### Inference on Webdataset Large-Scale Data

```Shell
bash run_inference.sh ${GPU_START_ID} ${Metric} ${Model_Path} ${Data_Path} ${Tars_Per_GPU} ${Num_GPU}
```
Parameters to note:

- `GPU_START_ID`: for large-scale score generation using multi-machines, specify the index of machines
- `Metric`: quality scoring metric for generation, select among `image_text_matching`, `object_detail_fulfillment`, `caption_text_quality`, `semantic_understanding`, `all`
- `Model_Path`: path to the mlm filter model checkpoint
- `Data_Path`: path to the webdataset image-text tars
- `Tars_Per_GPU`: the number of webdataset image-text tars for a single-gpu to inference on
- `Num_GPU`: the number of GPUs for one machine, e.g. 1, 8, 16

## Fine-Tuning MLM as Data Filter

1. Prepare data

Please download the [50k multimodal instructions](https://huggingface.co/datasets/weizhiwang/mlm_filter_instructions) and save it to `./data/mlm_filter_instruct_50k_gpt4v_cc12m_4k.json`.

Please download the images from constituting datasets:

- COCO: [train2017](http://images.cocodataset.org/zips/train2017.zip)
- GQA: [images](https://downloads.cs.stanford.edu/nlp/data/gqa/images.zip)
- OCR-VQA: [
ocr_vqa_images_llava_v15.zip](https://huggingface.co/datasets/weizhiwang/llava_v15_instruction_images/resolve/main/ocr_vqa_images_llava_v15.zip).
- TextVQA: [train_val_images](https://dl.fbaipublicfiles.com/textvqa/images/train_val_images.zip)
- VisualGenome: [part1](https://cs.stanford.edu/people/rak248/VG_100K_2/images.zip), [part2](https://cs.stanford.edu/people/rak248/VG_100K_2/images2.zip)
- CC12M: ```unzip images.zip -C data/images```, the images are available at [Huggingface Data Repo](https://huggingface.co/datasets/weizhiwang/mlm_filter_instructions).

After downloading all of them, organize the data as follows in `./data/images`,

```
├── coco
│   └── train2017
├── gqa
│   └── images
├── ocr_vqa
│   └── images
├── textvqa
│   └── train_images
└── vg
│   ├── VG_100K
│   └── VG_100K_2
└── cc12m
```

OCR-VQA are repacked by ourselves to ensure there is no failed-to-download images which are included in LLaVA-v1.5-665k instruction dataset.

2. Start training!

Please refer to [LLaVA-Unified](https://github.com/Victorwz/LLaVA-Unified) for more fine-tuning guidance.

Training script with DeepSpeed ZeRO-3: [`LLaVA_Unified/scripts/mlm_filter/finetune.sh`](https://github.com/Victorwz/LLaVA-Unified/blob/main/scripts/mlm_filter/finetune.sh).

## Our Best CLIP Model on DataComp-Medium
We also open-sourced our pre-trained CLIP-ViT-B/32 checkppint under the DataComp-Medium Benchmark Controlled Setting in [weizhiwang/clip_datacomp_medium_itm_th_66_AND_odf_th_20_gpt4v](https://huggingface.co/weizhiwang/clip_datacomp_medium_itm_th_66_AND_odf_th_20_gpt4v). Our best model is trianed on the data filtered by both the ITM and ODF Quality Scores.

## License
MIT License

## Contacts
For any question or issue, please feel free to contact [weizhiwang@ucsb.edu]() or submit github issues.

## Citation

Please cite our paper if you find this repository interesting or helpful in your research:
```bibtex
@article{mlm-filter,
    title={Finetuned Multimodal Language Models Are High-Quality Image-Text Data Filters}, 
    author={Wang, Weizhi and Mrini, Khalil and Yang, Linjie and Kumar, Sateesh and Tian, Yu and Yan, Xifeng and Wang, Heng},
    publisher={arXiv preprint arXiv:2403.02677},
    year={2024},
}
```

## Credits

MLM-Filter is developed based on
- [Vicuna](https://github.com/lm-sys/FastChat): foudation language model for LLaVA
- [LLaVA](https://github.com/haotian-liu/LLaVA): the codebase for fine-tuning LLaVA as image-text data filters
- [DataComp](https://github.com/mlfoundations/datacomp): the codebase for data filtering and CLIP pre-training

MLM_Filter

Download

Today's picks

Was this content helpful?

Recent Feedback

MLM_Filter

Download

Today's picks

Was this content helpful?

Recent Feedback

Download README