| # NOTE: for minhash and duplicates, using aria2c results in status-429 | |
| # (too many requests). So we use plain wget | |
| # Similarly for docs and quality signals, files were not completely | |
| # downloaded due to rate-limit by RPv2 servers (data.together.xyz) | |
| CC_SNAPSHOT="2023-14" | |
| LANG="en" | |
| BASE_URL="https://data.together.xyz/redpajama-data-v2/v1.0.0" | |
| BASE_DIR="/mnt/weka/peacock/enfm-dataprocessing/RedPajamaV2/data" | |
| listings_tag="${LANG}-${CC_SNAPSHOT}-head_middle" | |
| mkdir -p ${BASE_DIR}/listings | |
| # wget "${BASE_URL}/listings/${listings_tag}.txt" -O "${BASE_DIR}/listings/${listings_tag}.txt" # already downloaded | |
| listings_file="${BASE_DIR}/listings/${listings_tag}.txt" | |
| # create aria2c input filenames | |
| # # 1. signals | |
| # component="quality_signals" | |
| # component_extension="signals.json.gz" | |
| # # 2. docs | |
| # component="documents" | |
| # component_extension="json.gz" | |
| # # 3. minhash | |
| # component="minhash" | |
| # component_extension="minhash.parquet" | |
| # # 4. duplicates | |
| # component="duplicates" | |
| # component_extension="duplicates.parquet" | |
| aria2c_inputfile="${BASE_DIR}/listings/${component}-${listings_tag}.txt" | |
| out_dir="${BASE_DIR}/${component}" # for downloaded files | |
| truncate -s 0 $aria2c_inputfile | |
| while read line; do | |
| url="${BASE_URL}/${component}/${line}.${component_extension}" | |
| dest="${line}.${component_extension}" # path relative to --dir option of aria2c | |
| echo $url >> $aria2c_inputfile | |
| echo " out=${dest}" >> $aria2c_inputfile | |
| done <"$listings_file" | |
| # download | |
| aria2c -x 8 --dir=$out_dir --input-file=$aria2c_inputfile --continue=true | |