Dataset Viewer
Auto-converted to Parquet Duplicate
token
stringlengths
1
191
frequency
int64
1
36.2M
document_frequency
int64
1
8.7M
.
36,174,594
8,701,009
the
28,806,701
7,712,172
,
25,825,435
7,411,743
of
15,196,930
6,562,023
a
13,702,107
6,064,770
and
13,122,701
6,429,164
to
12,127,360
5,845,454
in
9,878,122
5,316,788
is
8,939,790
4,795,058
-
6,373,461
3,088,206
for
5,321,415
3,437,088
)
4,395,632
2,547,869
(
4,324,869
2,531,530
or
4,169,070
2,616,359
you
3,941,761
1,923,004
that
3,748,118
2,661,625
are
3,531,138
2,360,009
on
3,494,157
2,447,652
:
3,359,912
1,948,863
it
3,294,158
2,194,825
as
3,255,966
2,208,177
'
3,239,255
1,830,193
with
3,179,974
2,360,533
your
2,866,471
1,464,771
from
2,434,940
1,820,648
by
2,434,775
1,848,150
be
2,362,774
1,766,775
1
2,334,703
1,669,629
an
2,258,920
1,705,192
can
2,237,531
1,620,619
s
2,184,884
1,514,221
this
2,096,951
1,646,089
at
1,902,581
1,447,281
2
1,837,052
1,363,463
have
1,672,944
1,281,686
/
1,520,601
736,503
not
1,480,258
1,170,388
was
1,449,642
963,905
if
1,434,341
1,081,130
$
1,395,225
488,914
i
1,325,221
655,955
will
1,304,543
941,979
3
1,273,742
990,712
one
1,266,761
1,017,605
which
1,182,074
987,319
more
1,153,733
937,951
has
1,134,764
929,358
but
1,063,876
922,293
all
1,029,647
858,168
when
1,010,387
839,552
also
993,681
859,848
they
993,013
731,522
;
989,817
518,444
may
954,224
742,912
most
930,354
785,122
other
920,284
788,671
about
911,564
723,699
time
825,100
616,333
4
814,903
636,509
than
798,568
643,761
there
791,838
668,116
up
787,078
640,061
used
763,870
590,257
their
751,896
593,375
5
742,653
559,997
use
731,669
578,627
how
708,386
553,347
new
706,591
530,501
first
705,163
572,529
he
702,778
520,295
some
690,744
588,644
what
662,146
541,704
its
662,086
536,341
we
646,112
454,360
into
642,054
528,319
two
635,905
525,013
%
619,432
311,518
out
603,209
504,734
so
598,096
515,341
name
594,823
297,834
do
589,102
480,125
who
581,582
479,828
year
572,073
410,550
these
568,748
495,134
people
564,618
447,370
his
555,129
383,454
t
554,698
425,454
years
552,647
415,081
only
550,527
479,010
like
550,257
460,630
any
549,180
467,863
many
538,283
467,399
after
532,335
447,726
get
528,497
428,302
water
526,680
304,485
10
513,415
395,492
per
513,221
301,214
between
509,343
416,822
over
501,814
426,508
such
498,319
428,661
End of preview. Expand in Data Studio

MsMarco vocabulary counts

This is the vocabulary of the document part of the MsMarco dataset. This vocabulary was obtained by normalizing and pretokenizing the vocabulary using the bert-base-uncased tokenizer. You can use this vocabulary to:

  1. Obtain probabilities of subparts of your corpus.
  2. Define useful tokenizer extensions without fitting a new tokenizer.
  3. Analyzing the semantic content of the corpus

The dataset consists of 1.85 million tokens with their associated frequency and document frequency. The dataset is already sorted by frequency, so taking the N top rows also gets you the N most frequent tokens.

Acknowledgments

Thanks Mixedbread AI for a GPU grant for research into small retrieval models.

Downloads last month
4