Awesome

RBDash v1.5

install

Clone this repository and navigate to RBDash folder

git clone https://github.com/rbdash.git
cd RBDash

Install Package

conda create -n rbdash python=3.10 -y
conda activate rbdash
pip install --upgrade pip
pip install -e .

Install additional packages for training cases

pip install ninja
pip install flash-attn --no-build-isolation

or install the specific version of flash_attn from the .whl file: If you have already downloaded the flash_attn wheel file, for example, flash_attn-2.5.8+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl, you can install it with the following command:

pip install flash_attn-2.5.8+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl

Upgrade to latest code base

git pull
pip uninstall transformers
pip install -e .

Pretrained Weights

We recommend users to download the pretrained weights from the following link OpenCLIP-ConvNeXt-L, InternViT-6B-448px-V1-5,and put them in model_zoo following Structure.

Structure

RBDASH
├── rbdash
├── scripts
├── model_zoo
│   ├── OpenAI
│   │   ├── InternViT-6B-448px-V1-5
│   │   ├── openclip-convnext-large-d-320-laion2B-s29B-b131K-ft-soup
│   │   ├── ...

Model Zoo

RBDash-v1.5

Evaluation

In RBDash, we evaluate models on MME.

MME

Download the data following the official instructions here.
Downloaded images to ./rbdash-Eval/MME/MME_Benchmark_release_version.
Downloaded and put the weights to ./models/RBDash-v1.5
inference and evaluate.

bash scripts/rbdash/eval/mme.sh