h5py>=3.8
numpy>=1.24
pandas>=2.0
scipy>=1.10
tabulate>=0.9
scikit-learn>=1.3
tqdm>=4.60
torch>=2.0
torchaudio>=2.0
torchvision>=0.15
transformers>=4.36
sentencepiece>=0.2
spacy>=2.3
PyYAML>=6.0
absl-py
kaggle
requests>=2.28
tables
fsspec[http]
dataclasses-json
jsonschema>=4.17
tensorboard
torchmetrics>=1.0
torchinfo
filelock
psutil
protobuf>=4.0
gpustat
rich~=12.4.4
packaging
retry

# required for TransfoXLTokenizer when using transformer_xl
sacremoses
sentencepiece

# requirement for various paged and 8-bit optimizers
bitsandbytes>=0.44.0

# new data format support
xlwt            # excel
xlrd            # excel
openpyxl        # excel
pyarrow>=14.0   # parquet
lxml            # html

# requirement for loading hugging face datasets
datasets
