accelerate
transformers
sentencepiece
pillow
numpy
torchvision
huggingface_hub
spaces
opencv-python
imageio
imageio-ffmpeg
einops
timm
av
flash-attn-3@https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl
git+https://github.com/huggingface/diffusers.git@main


Common
                # Configuration system.
lpips>=0.1.4                    # LPIPS model for VAE training

# Dataloading
bson>=0.5.10                    # bson for Image dataloading
tensorflow>=2.16.1              # Video dataloading
opencv-python>=4.9.0.80         # OpenCV
sentencepiece>=0.2.0            # For Text Encoder

# Modeling
rotary-embedding-torch>=0.5.3   # Rotary positional embedding
transformers>=4.38.2            # Transformers
torchvision>=0.19.0             # Torchvision
tiktoken>=0.7.0                 # Tiktoken for generation
transformers_stream_generator>=0.0.5 #LLM generation support

# Metrics
torchmetrics>=1.3.2             # Core module for metric
pycocotools>=2.0.7              # COCO-related
torch-fidelity>=0.3.0           # FID-related

# Experiment Tracking
moviepy>=1.0.3                  # WandB Logging Image & Video
imageio>=2.34.0                 # WandB Logging Image & Video
tabulate>=0.9.0                 # Logging Table
deepdiff>=7.0.1                 # Find difference of config

# Testing
parameterized>=0.9.0            # Define multiple tests through decorators.

# Notebook
mediapy>=1.2.0                  # Notebook Visualization

# DevOPs
black >= 24             # Code formatting
flake8 >= 7              # Code style
isort >= 5               # Import sorting
pre-commit>=3.7.0               # Pre-commit hooks
expecttest>=0.2.1               # Pytorch dist tests
hypothesis>=6.100.1             # Fix randomness
av>=12.0.0