prepare_upload.sh

#!/bin/bash
#
# Prepare taco_dataset for HuggingFace upload.
# Splits files > 50GB into 49GB chunks and generates a manifest.
#
# Usage: ./prepare_upload.sh [REPO_ID]
#   REPO_ID defaults to mzhobro/taco_dataset
#
# After running, upload with:
#   huggingface-cli upload <REPO_ID> . --repo-type dataset
#
set -euo pipefail
cd "$(dirname "$0")"

THRESHOLD=$((50 * 1024 * 1024 * 1024))  # 50 GB
CHUNK="49G"
BACKUP_DIR="../taco_originals_backup"

echo "=== Finding files > 50GB ==="
LARGE_FILES=()
while IFS= read -r -d '' f; do
    size=$(stat -c%s "$f")
    if [ "$size" -gt "$THRESHOLD" ]; then
        LARGE_FILES+=("$f")
    fi
done < <(find . -type f -name '*.zip' -print0)

if [ ${#LARGE_FILES[@]} -eq 0 ]; then
    echo "No files exceed 50GB. Nothing to split."
else
    echo "Found ${#LARGE_FILES[@]} files to split:"
    for f in "${LARGE_FILES[@]}"; do
        echo "  $f ($(du -h "$f" | cut -f1))"
    done
    echo ""

    mkdir -p "$BACKUP_DIR"
    for f in "${LARGE_FILES[@]}"; do
        echo "[SPLIT] $f"
        split -b "$CHUNK" -d --additional-suffix=.part "$f" "${f}."
        echo "  Parts created:"
        ls -lh "${f}".*.part
        # Move original to backup (instant on same filesystem)
        backup_path="$BACKUP_DIR/$f"
        mkdir -p "$(dirname "$backup_path")"
        mv "$f" "$backup_path"
        echo "  Original moved to $backup_path"
        echo ""
    done
fi

echo "=== Generating manifest ==="
find . -type f \( -name '*.zip' -o -name '*.part' -o -name '*.md' \) | sort | while read -r f; do
    size=$(stat -c%s "$f")
    echo "$size  $f"
done > MANIFEST.txt
echo "Wrote MANIFEST.txt ($(wc -l < MANIFEST.txt) files)"

echo ""
echo "=== Done ==="
echo "To upload:"
echo "  huggingface-cli upload ${1:-mzhobro/taco_dataset} . --repo-type dataset"