The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
repo: string
at_ms: int64
reason: string
removed: list<item: struct<path: string, rows: int64, bytes: int64, digest: string>>
child 0, item: struct<path: string, rows: int64, bytes: int64, digest: string>
child 0, path: string
child 1, rows: int64
child 2, bytes: int64
child 3, digest: string
rewritten: list<item: string>
child 0, item: string
key: string
added_ms: int64
role: string
key_id: string
to
{'role': Value('string'), 'key': Value('string'), 'key_id': Value('string'), 'added_ms': Value('int64')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
repo: string
at_ms: int64
reason: string
removed: list<item: struct<path: string, rows: int64, bytes: int64, digest: string>>
child 0, item: struct<path: string, rows: int64, bytes: int64, digest: string>
child 0, path: string
child 1, rows: int64
child 2, bytes: int64
child 3, digest: string
rewritten: list<item: string>
child 0, item: string
key: string
added_ms: int64
role: string
key_id: string
to
{'role': Value('string'), 'key': Value('string'), 'key_id': Value('string'), 'added_ms': Value('int64')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
umi-meta
Reference data for umi, the crawler behind the open-index/* datasets. Everything here is a file that umi reads at runtime, published byte for byte as the crawler ran it, so that anyone can reproduce a umi key without reading Rust.
Why this repo exists
A crawler's published output is only reproducible if the rules that produced it are published too. A URL key in an open-index dataset is a fingerprint of a canonical URL, and the canonical form depends on which query parameters were treated as tracking noise. If that list lives only as a constant in a source tree then reproducing a key means reading Rust, and most people who want to check our work are not going to.
So the list is a file, the file is compiled into the binary with include_str!, and the same file is published here. There is no transcription step that can drift.
Contents
canon/tracking-params.txt
The parameters removed by step 10 of URL canonicalisation, specified in docs/spec/11-extraction-and-dedup.md section 11.2.
Syntax, one rule per line, # starts a comment, comparison is case insensitive:
| Form | Meaning |
|---|---|
name |
remove this parameter whenever it appears |
prefix* |
remove any parameter whose name starts with prefix |
name? |
remove only when the value looks like a session token |
The third form is the interesting one. sid is a session token on plenty of sites and a product id on plenty of others, and the two mistakes are not symmetric. Leaving a session parameter in costs one duplicate row, which is noisy but recoverable. Removing a content parameter merges two distinct pages into one key, and the page that loses is never crawled, silently. So anything ambiguous is judged by its value: 16 or more characters, alphanumeric plus dash and underscore throughout, and at least one digit.
Adding a line to this file changes every key derived from a URL carrying that parameter. That makes it a canonicalisation version bump and not a patch, which is why the version string canon/1 is recorded in every segment header and every published manifest.
Reproducing a umi URL key
A URL key is the first 10 bytes of the blake3 digest of the canonical URL. Eighty bits gives 0.004 expected collisions at 100 billion URLs.
import blake3 # pip install blake3
def url_key(canonical_url: str) -> str:
return blake3.blake3(canonical_url.encode()).digest()[:10].hex()
Getting from a raw URL to the canonical form is the harder half, and the twelve steps are written out in the spec. Two of them are worth flagging because they are the opposite of what most canonicalisers do:
- Query parameters are not sorted and repeated names are not deduplicated. Both are semantically live on real sites often enough that the breakage outweighs the dedup win.
- Trailing slashes are left exactly as found.
/aand/a/are different keys.
Measured effect
Over 1.6 million URLs from open-index/ccrawl-urls (CC-MAIN-2026-25, sampled across ten partitions, 38099 hosts):
| rewritten by canonicalisation | 1.10% |
| query parameters removed | 0.80% |
path parameters removed (;jsessionid= and friends) |
0.27% |
| collapsed onto an existing key | 1.34% |
| rejected as not crawlable | 5 of 1600000 |
License
Apache 2.0, same as umi.
- Downloads last month
- 98