in case anyone tried to use this, heres the missing build and dependies
#3
by nickhor78 - opened
# 1. Create and activate clean Conda environment
conda create -n action python=3.10 -y
conda activate action
# 2. Install build dependencies and target PyTorch 2.4.1
pip install torch==2.4.1 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
# 3. Install OpenGVLab vision stack dependencies
pip install transformers==4.40.1 av imageio decord opencv-python sentencepiece==0.1.99 protobuf einops timm
# 4. Download and install the pre-compiled FlashAttention wheel
## NOTE: MUST go to cd ~ dir, cannot in /mnt/d/...
wget https://github.com/Dao-AILab/flash-attention/releases/download/v2.6.3/flash_attn-2.6.3+cu123torch2.4cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
pip install flash_attn-2.6.3+cu123torch2.4cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
# 5 8-bit quantisation (dont work, sadly)
pip install accelerate bitsandbytes