upload_to_hf.sh

#!/bin/bash
# After data generation completes, upload the parquet shards + frames to HF.
# Run with HF_TOKEN env var set, e.g.:
#   HF_TOKEN=hf_xxx OUT_DIR=./procthor_data REPO=whab13/prismvla-procthor-data bash upload_to_hf.sh

set -e
OUT_DIR="${OUT_DIR:-./procthor_data}"
REPO="${REPO:-whab13/prismvla-procthor-data}"

if [ -z "$HF_TOKEN" ]; then
  echo "Set HF_TOKEN first, e.g.: HF_TOKEN=hf_xxx bash upload_to_hf.sh"
  exit 1
fi

source .venv/bin/activate
pip install -q huggingface_hub

# Use repo type DATASET
python - <<PY
import os
from huggingface_hub import HfApi, create_repo
api = HfApi(token="$HF_TOKEN")
repo = "$REPO"
try:
    create_repo(repo, repo_type="dataset", exist_ok=True, token="$HF_TOKEN")
except Exception as e:
    print("create_repo:", e)

# Upload the whole dir (parquet shards + frames/ subdir + manifest + logs)
print(f"Uploading $OUT_DIR -> $REPO  (this will take a while...)")
api.upload_large_folder(
    folder_path="$OUT_DIR",
    repo_id="$REPO",
    repo_type="dataset",
)
print("DONE.")
PY