Datasets:

danwil
/

owt-ngrams

tokens sequencelengths 1 6	freq uint32 0 332M	n uint8 1 6	tokens_str stringlengths 1 66	tok5 int32 -1 50.3k	tok4 int32 -1 50.3k	tok3 int32 -1 50.3k	tok2 int32 -1 50.3k	tok1 int32 -1 50.3k	tok0 int32 0 50.3k	tok0_str stringlengths 1 66
[ 0 ]	5,487,308	1	!	-1	-1	-1	-1	-1	0	!
[ 1 ]	13,327,355	1	"	-1	-1	-1	-1	-1	1	"
[ 2 ]	453,372	1	#	-1	-1	-1	-1	-1	2	#
[ 3 ]	519,325	1	$	-1	-1	-1	-1	-1	3	$
[ 4 ]	3,917,074	1	%	-1	-1	-1	-1	-1	4	%
[ 5 ]	668,676	1	&	-1	-1	-1	-1	-1	5	&
[ 6 ]	4,899,625	1	'	-1	-1	-1	-1	-1	6	'
[ 7 ]	2,959,408	1	(	-1	-1	-1	-1	-1	7	(
[ 8 ]	19,473,764	1	)	-1	-1	-1	-1	-1	8	)
[ 9 ]	788,274	1	*	-1	-1	-1	-1	-1	9	*
[ 10 ]	849,945	1	+	-1	-1	-1	-1	-1	10	+
[ 11 ]	331,770,469	1	,	-1	-1	-1	-1	-1	11	,
[ 12 ]	73,115,588	1	-	-1	-1	-1	-1	-1	12	-
[ 13 ]	317,302,752	1	.	-1	-1	-1	-1	-1	13	.
[ 14 ]	12,120,103	1	/	-1	-1	-1	-1	-1	14	/
[ 15 ]	2,955,772	1	0	-1	-1	-1	-1	-1	15	0
[ 16 ]	5,978,432	1	1	-1	-1	-1	-1	-1	16	1
[ 17 ]	5,073,915	1	2	-1	-1	-1	-1	-1	17	2
[ 18 ]	3,895,993	1	3	-1	-1	-1	-1	-1	18	3
[ 19 ]	3,167,346	1	4	-1	-1	-1	-1	-1	19	4
[ 20 ]	3,560,767	1	5	-1	-1	-1	-1	-1	20	5
[ 21 ]	2,358,514	1	6	-1	-1	-1	-1	-1	21	6
[ 22 ]	2,189,522	1	7	-1	-1	-1	-1	-1	22	7
[ 23 ]	2,258,639	1	8	-1	-1	-1	-1	-1	23	8
[ 24 ]	1,938,115	1	9	-1	-1	-1	-1	-1	24	9
[ 25 ]	33,924,449	1	:	-1	-1	-1	-1	-1	25	:
[ 26 ]	7,581,302	1	;	-1	-1	-1	-1	-1	26	;
[ 27 ]	307,115	1	<	-1	-1	-1	-1	-1	27	<
[ 28 ]	956,882	1	=	-1	-1	-1	-1	-1	28	=
[ 29 ]	988,348	1	>	-1	-1	-1	-1	-1	29	>
[ 30 ]	12,128,614	1	?	-1	-1	-1	-1	-1	30	?
[ 31 ]	761,206	1	@	-1	-1	-1	-1	-1	31	@
[ 32 ]	5,493,307	1	A	-1	-1	-1	-1	-1	32	A
[ 33 ]	2,692,962	1	B	-1	-1	-1	-1	-1	33	B
[ 34 ]	3,055,331	1	C	-1	-1	-1	-1	-1	34	C
[ 35 ]	2,874,327	1	D	-1	-1	-1	-1	-1	35	D
[ 36 ]	1,399,952	1	E	-1	-1	-1	-1	-1	36	E
[ 37 ]	2,110,413	1	F	-1	-1	-1	-1	-1	37	F
[ 38 ]	2,001,010	1	G	-1	-1	-1	-1	-1	38	G
[ 39 ]	1,862,554	1	H	-1	-1	-1	-1	-1	39	H
[ 40 ]	8,556,245	1	I	-1	-1	-1	-1	-1	40	I
[ 41 ]	1,311,763	1	J	-1	-1	-1	-1	-1	41	J
[ 42 ]	2,000,100	1	K	-1	-1	-1	-1	-1	42	K
[ 43 ]	1,855,725	1	L	-1	-1	-1	-1	-1	43	L
[ 44 ]	2,451,631	1	M	-1	-1	-1	-1	-1	44	M
[ 45 ]	1,762,596	1	N	-1	-1	-1	-1	-1	45	N
[ 46 ]	1,372,599	1	O	-1	-1	-1	-1	-1	46	O
[ 47 ]	2,107,505	1	P	-1	-1	-1	-1	-1	47	P
[ 48 ]	696,484	1	Q	-1	-1	-1	-1	-1	48	Q
[ 49 ]	2,174,567	1	R	-1	-1	-1	-1	-1	49	R
[ 50 ]	5,586,036	1	S	-1	-1	-1	-1	-1	50	S
[ 51 ]	2,116,146	1	T	-1	-1	-1	-1	-1	51	T
[ 52 ]	1,136,509	1	U	-1	-1	-1	-1	-1	52	U
[ 53 ]	1,401,815	1	V	-1	-1	-1	-1	-1	53	V
[ 54 ]	1,481,801	1	W	-1	-1	-1	-1	-1	54	W
[ 55 ]	884,765	1	X	-1	-1	-1	-1	-1	55	X
[ 56 ]	1,045,843	1	Y	-1	-1	-1	-1	-1	56	Y
[ 57 ]	861,450	1	Z	-1	-1	-1	-1	-1	57	Z
[ 58 ]	1,684,351	1	[	-1	-1	-1	-1	-1	58	[
[ 59 ]	861,769	1	\	-1	-1	-1	-1	-1	59	\
[ 60 ]	8,698,958	1	]	-1	-1	-1	-1	-1	60	]
[ 61 ]	136,885	1	^	-1	-1	-1	-1	-1	61	^
[ 62 ]	4,129,741	1	_	-1	-1	-1	-1	-1	62	_
[ 63 ]	125,031	1	`	-1	-1	-1	-1	-1	63	`
[ 64 ]	5,567,006	1	a	-1	-1	-1	-1	-1	64	a
[ 65 ]	2,184,387	1	b	-1	-1	-1	-1	-1	65	b
[ 66 ]	2,210,917	1	c	-1	-1	-1	-1	-1	66	c
[ 67 ]	3,317,323	1	d	-1	-1	-1	-1	-1	67	d
[ 68 ]	2,643,819	1	e	-1	-1	-1	-1	-1	68	e
[ 69 ]	1,986,830	1	f	-1	-1	-1	-1	-1	69	f
[ 70 ]	1,822,512	1	g	-1	-1	-1	-1	-1	70	g
[ 71 ]	2,085,684	1	h	-1	-1	-1	-1	-1	71	h
[ 72 ]	3,977,354	1	i	-1	-1	-1	-1	-1	72	i
[ 73 ]	1,385,829	1	j	-1	-1	-1	-1	-1	73	j
[ 74 ]	1,975,873	1	k	-1	-1	-1	-1	-1	74	k
[ 75 ]	1,878,104	1	l	-1	-1	-1	-1	-1	75	l
[ 76 ]	5,278,088	1	m	-1	-1	-1	-1	-1	76	m
[ 77 ]	1,990,360	1	n	-1	-1	-1	-1	-1	77	n
[ 78 ]	3,399,287	1	o	-1	-1	-1	-1	-1	78	o
[ 79 ]	1,698,336	1	p	-1	-1	-1	-1	-1	79	p
[ 80 ]	532,952	1	q	-1	-1	-1	-1	-1	80	q
[ 81 ]	1,718,866	1	r	-1	-1	-1	-1	-1	81	r
[ 82 ]	46,501,587	1	s	-1	-1	-1	-1	-1	82	s
[ 83 ]	15,084,616	1	t	-1	-1	-1	-1	-1	83	t
[ 84 ]	1,776,410	1	u	-1	-1	-1	-1	-1	84	u
[ 85 ]	1,492,903	1	v	-1	-1	-1	-1	-1	85	v
[ 86 ]	1,261,040	1	w	-1	-1	-1	-1	-1	86	w
[ 87 ]	1,993,162	1	x	-1	-1	-1	-1	-1	87	x
[ 88 ]	3,326,055	1	y	-1	-1	-1	-1	-1	88	y
[ 89 ]	1,838,137	1	z	-1	-1	-1	-1	-1	89	z
[ 90 ]	261,610	1	{	-1	-1	-1	-1	-1	90	{
[ 91 ]	577,959	1	\|	-1	-1	-1	-1	-1	91	\|
[ 92 ]	410,243	1	}	-1	-1	-1	-1	-1	92	}
[ 93 ]	112,424	1	~	-1	-1	-1	-1	-1	93	~
[ 94 ]	187,522	1	�	-1	-1	-1	-1	-1	94	�
[ 95 ]	114,158	1	�	-1	-1	-1	-1	-1	95	�
[ 96 ]	137,071	1	�	-1	-1	-1	-1	-1	96	�
[ 97 ]	165,632	1	�	-1	-1	-1	-1	-1	97	�
[ 98 ]	161,228	1	�	-1	-1	-1	-1	-1	98	�
[ 99 ]	93,769	1	�	-1	-1	-1	-1	-1	99	�

End of preview. Expand in Data Studio

Dataset Card for OpenWebText n-grams

Dataset Summary

This dataset contains 246K of the most common token-based (GPT-2/GPT-3) n-grams (n=1 to n=6), in the OpenWebText (OWT) dataset.

For convenient searching, it provides full tokens/strings, as well as per-position tokens/strings.

Usage

Generally, this dataset allows identifying the most common n-grams in a text corpus.

When researching LLMs tokenized similarly to GPT-2/GPT-3, it allows:

Constructing intermediate vectors spanning the most common short phrases (n-grams), e.g. for similarity sampling.
Fast searches for common phrases containing particular tokens or substrings (and in particular sequence positions).
Showing the effects of training set n-gram frequency.

The authors (Thomas Dooms and Dan Wilhelm) used this dataset to show that sparse auto-encoders are biased toward reconstructing the most common n-grams.

Loading the Dataset

We recommend you convert the dataset to a Pandas DataFrame for easy querying:

from datasets import load_dataset

ngrams = load_dataset('danwil/owt-ngrams')['train'].to_pandas()

Below, we list the number of n-grams and their count/frequency in the original ~9B-token OWT corpus.

We include all individual tokens (1-grams).
Note that if an n-gram occurs >N times, then every contiguous subsequence must also occur >N times.

	total	n=1	n=2	n=3	n=4	n=5	n=6
owt_1-6grams_246k	245831	50257	58302	44560	32831	13566	12495
count in OWT		>= 0	>= 10000	>= 10000	> 5000	> 5000	> 2000

Point of Contact: Dan Wilhelm

Downloads last month: 33

Size of downloaded dataset files:

6.21 MB

Size of the auto-converted Parquet files:

6.21 MB

Number of rows:

245,831

Dataset Card for OpenWebText n-grams

Dataset Summary

Usage

Loading the Dataset

Contents