Category	Count	%	Mean Tokens	Median	Std	Min	Max	Total Tokens
math	184,369	36.92%	19,409.8	18,968	6,162.1	1,149	35,745	3,578,569,601
code	169,849	34.01%	7,407.2	4,482	7,152.3	268	36,577	1,258,098,243
instruction-following	80,949	16.21%	2,237.1	1,161	3,050.4	38	34,005	181,093,270
science	25,456	5.10%	8,920.8	7,725	5,896.2	614	34,041	227,088,132
safety	19,489	3.90%	744.9	678	386.6	83	11,841	14,517,816
general-chat	18,121	3.63%	3,052.2	2,023	2,964.9	32	30,771	55,309,560
structured-data	1,145	0.23%	1,704.2	1,210	1,603.4	198	13,211	1,951,280

Token Count Statistics

Dataset to Category Mapping

Category	Original Dataset	Count	%	Mean Tokens
code	Dolci-Think-Python	121,923	24.41%	4,404.6
code	Nemotron Post-Training	27,328	5.47%	13,273.3
code	OpenThoughts3 (code)	20,598	4.12%	17,396.7
general-chat	WildChat (Tulu v3.9)	9,726	1.95%	3,118.7
general-chat	WildChat	7,596	1.52%	3,126.4
general-chat	OpenAssistant (OASST1)	799	0.16%	1,537.6
instruction-following	Dolci-Precise-IF	53,699	10.75%	1,473.0
instruction-following	SYNTHETIC-2 (IF/QwQ)	27,250	5.46%	3,742.9
math	OpenThoughts3 (math)	184,369	36.92%	19,409.8
safety	WildJailbreak	9,182	1.84%	649.7
safety	WildGuardMix	8,134	1.63%	811.2
safety	CoCoNot	2,173	0.44%	899.5
science	OpenThoughts3 (science)	25,456	5.10%	8,920.8
structured-data	TableGPT	1,145	0.23%	1,704.2