google-genai>=1.0
datasets>=2.18
huggingface_hub>=0.20
lancedb>=0.6
numpy>=1.24
pyarrow>=14.0
# For benchmark only:
scipy>=1.11
