# Copy to .env. Never commit the real token.
HF_TOKEN=hf_replace_me

# Optional experiment tracking. Training works without it.
WANDB_MODE=offline