Datasets:

---

allenai

/

dolma3_pool

like 32

Follow

Ai2 5.41k

I noticed a very rare few files in this dataset appear to be incomplete, ending in a line that stop in the middle rather than completing as a valid JSON string.

One such example is dolma3_pool/data/common_crawl-software-0019/shard_00000152.jsonl.zst for which the final line (line 38290) which ends abruptly in the text section of the json line.

This causes an error when trying to read and parse the file's json, e.g.:

pd.read_json("~/Downloads/shard_00000152.jsonl.zst", lines=True)
# ValueError: Unmatched ''"' when when decoding 'string'

More detailed error information can be obtained by parsing line-by-line with a try-catch:

from pathlib import path

import zstandard as zstd
import json
import io
from tqdm.auto import tqdm

# NOTE: Initialize p_list with a list of Path objects pointing to the files.
for p in p_list:
    print(p)
    with open(p, "rb") as f:
        dctx = zstd.ZstdDecompressor()
        with dctx.stream_reader(f) as reader:
            text_reader = io.TextIOWrapper(reader, encoding="utf-8")
            for i, line in enumerate(tqdm(text_reader)):
                line = line.rstrip("\n")
                if not line:
                    continue
                try:
                    json.loads(line)
                except json.JSONDecodeError as e:
                    print(f"Line {i + 1}: {e}")
                    # Example: "Line 35254: Unterminated string starting at: line 1 column 1120 (char 1119)"
                    print(f"Raw: {repr(line)}")

In total I had to skip all of the following incomplete files when reading the common-crawl subset of the olmo3 pool dataset:

common_crawl-software-0019/shard_00000152.jsonl.zst
common_crawl-travel_and_tourism-0015/shard_00000312.jsonl.zst
common_crawl-social_life-0015/shard_00000129.jsonl.zst
common_crawl-transportation-0019/shard_00000094.jsonl.zst
common_crawl-transportation-0019/shard_00000092.jsonl.zst
common_crawl-travel_and_tourism-0007/shard_00000378.jsonl.zst

Given that all of these files were recently uploaded after discussion #3, I think there may have been some issue in that particular upload, and it is quite possible that this issue does not affect the other portions of the dataset, but it may be useful to check.