messages stringlengths 154 534 | tools stringclasses 1
value | src stringclasses 1
value | meta stringclasses 304
values |
|---|---|---|---|
[{"role": "user", "content": "Count the 'd's in the word 'denominator'."}, {"role": "assistant", "content": "<think>denominator: d1* e2 n3 o4 m5 i6 n7 a8 t9 o10 r11. 'd' is at position 1, so 1.</think>\nThe letter 'd' appears 1 time in 'denominator'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "user", "content": "Are 'tea' and 'eat' anagrams of each other?"}, {"role": "assistant", "content": "<think>Sorted letters of 'tea': aet. Sorted letters of 'eat': aet. They match.</think>\nYes."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "anagram", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What is the fifth letter of 'assignment'?"}, {"role": "assistant", "content": "<think>'assignment': a1 s2 s3 i4 g5 n6 m7 e8 n9 t10. Position 5 is g.</think>\ng"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "user", "content": "How many letters are there between M and Z in the alphabet?"}, {"role": "assistant", "content": "<think>M is 13 and Z is 26. The letters strictly between them: 26 - 13 - 1 = 12.</think>\nThere are 12 between M and Z."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "Spell 'maximum' with hyphens between the letters."}, {"role": "assistant", "content": "<think>'maximum' has 7 letters: m1 a2 x3 i4 m5 u6 m7.</think>\nm-a-x-i-m-u-m"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "spell", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "What is the next letter after Y?"}, {"role": "assistant", "content": "<think>Y is letter 25, so the next is letter 26, which is Z.</think>\nZ"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "after", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Is the letter W a vowel?"}, {"role": "assistant", "content": "<think>The vowels are A, E, I, O, U. W is not in that list.</think>\nA consonant."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "vowel", "kid": 0} | |
[{"role": "user", "content": "Spell the word 'integration' letter by letter, separated by hyphens."}, {"role": "assistant", "content": "<think>'integration' has 11 letters: i1 n2 t3 e4 g5 r6 a7 t8 i9 o10 n11.</think>\ni-n-t-e-g-r-a-t-i-o-n"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "spell", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "How many letters are there between P and U in the alphabet?"}, {"role": "assistant", "content": "<think>P is 16 and U is 21. The letters strictly between them: 21 - 16 - 1 = 4.</think>\n4... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Reverse the letters of 'romantic'."}, {"role": "assistant", "content": "<think>'romantic' has 8 letters: r1 o2 m3 a4 n5 t6 i7 c8. Reading from 8 down to 1: c i t n a m o r.</think>\ncitna... | alphabet_letters | {"family": "alphabet_letters", "kind": "reverse", "kid": 0} | |
[{"role": "user", "content": "Is 'curl' a palindrome?"}, {"role": "assistant", "content": "<think>'curl' backwards is 'lruc', which differs from 'curl'.</think>\nNo."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "palindrome", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "What position is the letter S in the alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... S is number 19.</think>\n19"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "position", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What is the third letter of 'aperture'?"}, {"role": "assistant", "content": "<think>'aperture': a1 p2 e3 r4 t5 u6 r7 e8. Position 3 is e.</think>\ne"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "How many letters are there between D and J in the alphabet?"}, {"role": "assistant", "content": "<think>D is 4 and J is 10. The letters strictly between them: 10 - 4 - 1 = 5.</think>\nThe... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "What letter follows d in the English alphabet?"}, {"role": "assistant", "content": "<think>D is letter 4, so the next is letter 5, which is E.</think>\ne"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "after", "kid": 0} | |
[{"role": "user", "content": "Count the vowels in the word 'mim'."}, {"role": "assistant", "content": "<think>mim: m1 i2* m3. Vowels: i2. Total 1.</think>\n'mim' has 1 vowel."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_vowels", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "What is the seventh letter of 'filling'?"}, {"role": "assistant", "content": "<think>'filling': f1 i2 l3 l4 i5 n6 g7. Position 7 is g.</think>\ng"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "Put the letters of 'communist' in alphabetical order, separated by spaces."}, {"role": "assistant", "content": "<think>Letters of 'communist': c o m m u n i s t. Sorted from a to z: c i m m n o s t u.... | alphabet_letters | {"family": "alphabet_letters", "kind": "sort_letters", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Which letter is two places after I?"}, {"role": "assistant", "content": "<think>I is letter 9; adding 2 gives 11, which is K.</think>\nK"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "skip", "kid": 0} | |
[{"role": "user", "content": "Which letter is in position 6 of the word 'narrower'?"}, {"role": "assistant", "content": "<think>'narrower': n1 a2 r3 r4 o5 w6 e7 r8. Position 6 is w.</think>\nw"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "Spell the word 'chose' letter by letter, separated by hyphens."}, {"role": "assistant", "content": "<think>'chose' has 5 letters: c1 h2 o3 s4 e5.</think>\nc-h-o-s-e"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "spell", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "What is letter 21 of the alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... the letter at number 21 is U.</think>\nU"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth", "kid": 0} | |
[{"role": "user", "content": "What letter comes five letters after C in the alphabet?"}, {"role": "assistant", "content": "<think>C is letter 3; adding 5 gives 8, which is H.</think>\nH"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "skip", "kid": 0} | |
[{"role": "user", "content": "How many letters are there between A and J in the alphabet?"}, {"role": "assistant", "content": "<think>A is 1 and J is 10. The letters strictly between them: 10 - 1 - 1 = 8.</think>\nThere are 8 between A and J."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What is letter 4 of the alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... the letter at number 4 is D.</think>\nD"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "Are 'lemon' and 'lelon' anagrams of each other?"}, {"role": "assistant", "content": "<think>Sorted letters of 'lemon': elmno. Sorted letters of 'lelon': ellno. They differ.</think>\nNo."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "anagram", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "Reverse the letters of 'unequal'."}, {"role": "assistant", "content": "<think>'unequal' has 7 letters: u1 n2 e3 q4 u5 a6 l7. Reading from 7 down to 1: l a u q e n u.<... | alphabet_letters | {"family": "alphabet_letters", "kind": "reverse", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "Write the 26 letters of the English alphabet."}, {"role": "assistant", "content": "<think>The English alphabet has 26 letters from A to Z, in order: A1 B2 C3 D4 E5 F6... | alphabet_letters | {"family": "alphabet_letters", "kind": "alphabet", "kid": 0} | |
[{"role": "user", "content": "Spell 'valued' with hyphens between the letters."}, {"role": "assistant", "content": "<think>'valued' has 6 letters: v1 a2 l3 u4 e5 d6.</think>\nv-a-l-u-e-d"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "spell", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "What are the first and last letters of 'extreme'?"}, {"role": "assistant", "content": "<think>'extreme' starts with e and ends with e.</think>\nFirst: e, last: e."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "first_last", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Give me all letters from A to Z. Use uppercase letters, separated by spaces."}, {"role": "assistant", "content": "<think>The English alphabet has 26 letters from A to Z, in order: A1 B2 C... | alphabet_letters | {"family": "alphabet_letters", "kind": "alphabet", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What letter sits between X and Z in the alphabet?"}, {"role": "assistant", "content": "<think>X is letter 24 and Z is letter 26, so the one between is letter 25: Y.</think>\nY"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Put the letters of 'quietly' in alphabetical order, separated by spaces."}, {"role": "assistant", "content": "<think>Letters of 'quietly': q u i e t l y. Sorted from a to z: e i l q t u y... | alphabet_letters | {"family": "alphabet_letters", "kind": "sort_letters", "kid": 0} | |
[{"role": "user", "content": "What is the second letter of 'cavity'?"}, {"role": "assistant", "content": "<think>'cavity': c1 a2 v3 i4 t5 y6. Position 2 is a.</think>\na"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "user", "content": "Write the word 'walk' backwards."}, {"role": "assistant", "content": "<think>'walk' has 4 letters: w1 a2 l3 k4. Reading from 4 down to 1: k l a w.</think>\nklaw"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "reverse", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "What letter sits between D and F in the alphabet?"}, {"role": "assistant", "content": "<think>D is letter 4 and F is letter 6, so the one between is letter 5: E.</think>\nE"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Write the alphabet. Use uppercase letters, separated by spaces."}, {"role": "assistant", "content": "<think>The English alphabet has 26 letters from A to Z, in order: A1 B2 C3 D4 E5 F6 G7... | alphabet_letters | {"family": "alphabet_letters", "kind": "alphabet", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Is 'sleep' a palindrome?"}, {"role": "assistant", "content": "<think>'sleep' backwards is 'peels', which differs from 'sleep'.</think>\nNo, 'sleep' backwards is 'peels'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "palindrome", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Put the letters of 'warmer' in alphabetical order, separated by spaces."}, {"role": "assistant", "content": "<think>Letters of 'warmer': w a r m e r. Sorted from a to z: a e m r r w.</thi... | alphabet_letters | {"family": "alphabet_letters", "kind": "sort_letters", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "Which letter is four places after H?"}, {"role": "assistant", "content": "<think>H is letter 8; adding 4 gives 12, which is L.</think>\nL"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "skip", "kid": 0} | |
[{"role": "user", "content": "Which letter comes right after X?"}, {"role": "assistant", "content": "<think>X is letter 24, so the next is letter 25, which is Y.</think>\nY"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "after", "kid": 0} | |
[{"role": "user", "content": "Spell 'clay' backwards."}, {"role": "assistant", "content": "<think>'clay' has 4 letters: c1 l2 a3 y4. Reading from 4 down to 1: y a l c.</think>\nyalc"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "reverse", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "How many letters are there between R and T in the alphabet?"}, {"role": "assistant", "content": "<think>R is 18 and T is 20. The letters strictly between them: 20 - 18 - 1 = 1.</think>\nThere i... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Count the 'c's in the word 'minimum'."}, {"role": "assistant", "content": "<think>minimum: m1 i2 n3 i4 m5 u6 m7. No 'c' appears, so 0.</think>\nThe letter 'c' appears 0 times in 'minimum'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "user", "content": "What are the first and last letters of 'extract'?"}, {"role": "assistant", "content": "<think>'extract' starts with e and ends with t.</think>\nFirst: e, last: t."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "first_last", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "Are 'state' and 'taste' anagrams of each other?"}, {"role": "assistant", "content": "<think>Sorted letters of 'state': aestt. Sorted letters of 'taste': aestt. They m... | alphabet_letters | {"family": "alphabet_letters", "kind": "anagram", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "Count the 'r's in the word 'burnt'."}, {"role": "assistant", "content": "<think>burnt: b1 u2 r3* n4 t5. 'r' is at position 3, so 1.</think>\nThe letter 'r' appears 1 time in 'burnt'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What are the first and last letters of 'relax'?"}, {"role": "assistant", "content": "<think>'relax' starts with r and ends with x.</think>\nFirst: r, last: x."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "first_last", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Which letter is number 26 in the English alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... the letter at number 26 is Z.</think>\nZ"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth", "kid": 0} | |
[{"role": "user", "content": "Spell the word 'regain' letter by letter, separated by hyphens."}, {"role": "assistant", "content": "<think>'regain' has 6 letters: r1 e2 g3 a4 i5 n6.</think>\nr-e-g-a-i-n"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "spell", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "How many letters are there between K and O in the alphabet?"}, {"role": "assistant", "content": "<think>K is 11 and O is 15. The letters strictly between them: 15 - 1... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "user", "content": "What is the length of 'divided' in letters?"}, {"role": "assistant", "content": "<think>d1 i2 v3 i4 d5 e6 d7. The last number is 7.</think>\n7"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "length", "kid": 0} | |
[{"role": "user", "content": "Is the letter N a vowel?"}, {"role": "assistant", "content": "<think>The vowels are A, E, I, O, U. N is not in that list.</think>\nA consonant."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "vowel", "kid": 0} | |
[{"role": "user", "content": "How many letters are in the word 'cod'?"}, {"role": "assistant", "content": "<think>c1 o2 d3. The last number is 3.</think>\n'cod' has 3 letters."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "length", "kid": 0} | |
[{"role": "user", "content": "What comes after p in the alphabet?"}, {"role": "assistant", "content": "<think>P is letter 16, so the next is letter 17, which is Q.</think>\nq"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "after", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Spell the word 'tennis' letter by letter, separated by hyphens."}, {"role": "assistant", "content": "<think>'tennis' has 6 letters: t1 e2 n3 n4 i5 s6.</think>\nt-e-n-n-i-s"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "spell", "kid": 0} | |
[{"role": "user", "content": "What is the fifth letter of 'individually'?"}, {"role": "assistant", "content": "<think>'individually': i1 n2 d3 i4 v5 i6 d7 u8 a9 l10 l11 y12. Position 5 is v.</think>\nv"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "user", "content": "What letter sits between T and V in the alphabet?"}, {"role": "assistant", "content": "<think>T is letter 20 and V is letter 22, so the one between is letter 21: U.</think>\nU"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "between", "kid": 0} | |
[{"role": "user", "content": "Which number is the letter Y in the English alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... Y is number 25.</think>\nY is letter number 25."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "position", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "List the English alphabet in order. Use uppercase letters, separated by commas."}, {"role": "assistant", "content": "<think>The English alphabet has 26 letters from A... | alphabet_letters | {"family": "alphabet_letters", "kind": "alphabet", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "What position is the letter H in the alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... H is number 8.</think>\n8"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "position", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "What is the 23rd letter of the alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... the letter at number 23 is W.</think>\nW"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "How many times does the letter 'w' appear in 'will'?"}, {"role": "assistant", "content": "<think>will: w1* i2 l3 l4. 'w' is at position 1, so 1.</think>\nThe letter 'w' appears 1 time in 'will'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "user", "content": "How many letters are there between N and V in the alphabet?"}, {"role": "assistant", "content": "<think>N is 14 and V is 22. The letters strictly between them: 22 - 14 - 1 = 7.</think>\nThere are 7 between N and V."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "How many letters are there between Q and S in the alphabet?"}, {"role": "assistant", "content": "<think>Q is 17 and S is 19. The letters strictly between them: 19 - 17 - 1 = 1.</think>\nT... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "How many y's are in 'nationality'?"}, {"role": "assistant", "content": "<think>nationality: n1 a2 t3 i4 o5 n6 a7 l8 i9 t10 y11*. 'y' is at position 11, so 1.</think>\nThe letter 'y' appea... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "How many letters are there between M and X in the alphabet?"}, {"role": "assistant", "content": "<think>M is 13 and X is 24. The letters strictly between them: 24 - 13 - 1 = 10.</think>\n... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "How many d's are in 'bedroom'?"}, {"role": "assistant", "content": "<think>bedroom: b1 e2 d3* r4 o5 o6 m7. 'd' is at position 3, so 1.</think>\nThe letter 'd' appears 1 time in 'bedroom'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Reverse the letters of 'weapon'."}, {"role": "assistant", "content": "<think>'weapon' has 6 letters: w1 e2 a3 p4 o5 n6. Reading from 6 down to 1: n o p a e w.</think>\nnopaew"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "reverse", "kid": 0} | |
[{"role": "user", "content": "Which letter is in position 1 of the word 'silk'?"}, {"role": "assistant", "content": "<think>'silk': s1 i2 l3 k4. Position 1 is s.</think>\ns"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "How many letters are in the word 'foil'?"}, {"role": "assistant", "content": "<think>f1 o2 i3 l4. The last number is 4.</think>\n'foil' has 4 letters."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "length", "kid": 0} | |
[{"role": "user", "content": "How many times does the letter 'k' appear in 'mask'? Number only, please."}, {"role": "assistant", "content": "<think>mask: m1 a2 s3 k4*. 'k' is at position 4, so 1.</think>\n1"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "How many c's are in 'gluten'?"}, {"role": "assistant", "content": "<think>gluten: g1 l2 u3 t4 e5 n6. No 'c' appears, so 0.</think>\nThe letter 'c' appears 0 times in 'gluten'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "How many letters are there between R and V in the alphabet?"}, {"role": "assistant", "content": "<think>R is 18 and V is 22. The letters strictly between them: 22 - 1... | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Which letter is number 25 in the English alphabet?"}, {"role": "assistant", "content": "<think>Counting A1 B2 C3 ... the letter at number 25 is Y.</think>\nY"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth", "kid": 0} | |
[{"role": "user", "content": "Which letter is in position 5 of the word 'speed'?"}, {"role": "assistant", "content": "<think>'speed': s1 p2 e3 e4 d5. Position 5 is d.</think>\nd"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "user", "content": "What letter sits between H and J in the alphabet?"}, {"role": "assistant", "content": "<think>H is letter 8 and J is letter 10, so the one between is letter 9: I.</think>\nI"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Put the letters of 'medieval' in alphabetical order, separated by spaces."}, {"role": "assistant", "content": "<think>Letters of 'medieval': m e d i e v a l. Sorted from a to z: a d e e i... | alphabet_letters | {"family": "alphabet_letters", "kind": "sort_letters", "kid": 0} | |
[{"role": "user", "content": "How many vowels are in 'climate'?"}, {"role": "assistant", "content": "<think>climate: c1 l2 i3* m4 a5* t6 e7*. Vowels: i3, a5, e7. Total 3.</think>\n'climate' has 3 vowels."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_vowels", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "Spell the word 'patch' letter by letter, separated by hyphens."}, {"role": "assistant", "content": "<think>'patch' has 5 letters: p1 a2 t3 c4 h5.</think>\np-a-t-c-h"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "spell", "kid": 0} | |
[{"role": "user", "content": "Is 'server' a palindrome?"}, {"role": "assistant", "content": "<think>'server' backwards is 'revres', which differs from 'server'.</think>\nNo, 'server' backwards is 'revres'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "palindrome", "kid": 0} | |
[{"role": "user", "content": "How many times does the letter 'y' appear in 'posit'?"}, {"role": "assistant", "content": "<think>posit: p1 o2 s3 i4 t5. No 'y' appears, so 0.</think>\nThe letter 'y' appears 0 times in 'posit'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "user", "content": "How many times does the letter 'z' appear in 'enzyme'? Give only the number, nothing else."}, {"role": "assistant", "content": "<think>enzyme: e1 n2 z3* y4 m5 e6. 'z' is at position 3, so 1.</think>\n1"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "user", "content": "How many letters are there between E and X in the alphabet?"}, {"role": "assistant", "content": "<think>E is 5 and X is 24. The letters strictly between them: 24 - 5 - 1 = 18.</think>\nThere are 18 between E and X."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_between", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What letter sits between C and E in the alphabet?"}, {"role": "assistant", "content": "<think>C is letter 3 and E is letter 5, so the one between is letter 4: D.</think>\nD"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "between", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "How many letters are in the word 'enrich'?"}, {"role": "assistant", "content": "<think>e1 n2 r3 i4 c5 h6. The last number is 6.</think>\n6"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "length", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Count the vowels in the word 'positional'."}, {"role": "assistant", "content": "<think>positional: p1 o2* s3 i4* t5 i6* o7* n8 a9* l10. Vowels: o2, i4, i6, o7, a9. Total 5.</think>\n5"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_vowels", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What is the second letter of 'foster'?"}, {"role": "assistant", "content": "<think>'foster': f1 o2 s3 t4 e5 r6. Position 2 is o.</think>\no"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "nth_letter", "kid": 0} | |
[{"role": "user", "content": "Sort the letters of 'ruling' in alphabetical order."}, {"role": "assistant", "content": "<think>Letters of 'ruling': r u l i n g. Sorted from a to z: g i l n r u.</think>\ng i l n r u"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "sort_letters", "kid": 0} | |
[{"role": "user", "content": "Are 'cried' and 'cidej' anagrams of each other?"}, {"role": "assistant", "content": "<think>Sorted letters of 'cried': cdeir. Sorted letters of 'cidej': cdeij. They differ.</think>\nNo."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "anagram", "kid": 0} | |
[{"role": "system", "content": "You are a helpful assistant. Follow the user's instructions exactly."}, {"role": "user", "content": "What comes after I in the alphabet?"}, {"role": "assistant", "content": "<think>I is letter 9, so the next is letter 10, which is J.</think>\nJ"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "after", "kid": 0} | |
[{"role": "user", "content": "How many times does the letter 'i' appear in 'lifting'?"}, {"role": "assistant", "content": "<think>lifting: l1 i2* f3 t4 i5* n6 g7. 'i' is at positions 2, 5, so 2.</think>\nThe letter 'i' appears 2 times in 'lifting'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "What letter sits between F and H in the alphabet?"}, {"role": "assistant", "content": "<think>F is letter 6 and H is letter 8, so the one between is letter 7: G.</think>\nG"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "between", "kid": 0} | |
[{"role": "user", "content": "Write the word 'motif' backwards."}, {"role": "assistant", "content": "<think>'motif' has 5 letters: m1 o2 t3 i4 f5. Reading from 5 down to 1: f i t o m.</think>\nfitom"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "reverse", "kid": 0} | |
[{"role": "user", "content": "Count the 'e's in the word 'harder'."}, {"role": "assistant", "content": "<think>harder: h1 a2 r3 d4 e5* r6. 'e' is at position 5, so 1.</think>\nThe letter 'e' appears 1 time in 'harder'."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} | |
[{"role": "user", "content": "Count the vowels in the word 'basement'."}, {"role": "assistant", "content": "<think>basement: b1 a2* s3 e4* m5 e6* n7 t8. Vowels: a2, e4, e6. Total 3.</think>\n'basement' has 3 vowels."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_vowels", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "Which letter is between V and X?"}, {"role": "assistant", "content": "<think>V is letter 22 and X is letter 24, so the one between is letter 23: W.</think>\nW"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "between", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a precise, analytical, and honest AI assistant."}, {"role": "user", "content": "Sort the letters of 'working' in alphabetical order."}, {"role": "assistant", "content": "<think>Letters of 'working': w o r k i n g. Sorted from a to z: g i k n o r w.</think>\ng i k n o r w"}... | alphabet_letters | {"family": "alphabet_letters", "kind": "sort_letters", "kid": 0} | |
[{"role": "system", "content": "You are an AI assistant. Answer exactly what was asked."}, {"role": "user", "content": "Is P a vowel or a consonant?"}, {"role": "assistant", "content": "<think>The vowels are A, E, I, O, U. P is not in that list.</think>\nA consonant."}] | alphabet_letters | {"family": "alphabet_letters", "kind": "vowel", "kid": 0} | |
[{"role": "system", "content": "You are Zenyx, a careful assistant that follows instructions exactly and checks its work."}, {"role": "user", "content": "How many a's are in 'sat'? Answer with only the number."}, {"role": "assistant", "content": "<think>sat: s1 a2* t3. 'a' is at position 2, so 1.</think>\n1"}] | alphabet_letters | {"family": "alphabet_letters", "kind": "count_letter", "kid": 0} |
Zenyx Instruct Exact
About 250 Million Rule-Based Exact-Instruction Rows in 33 Categories
Every Answer Computed by Code โข Code Answers Executed โข Think-Style, Multi-Turn โข Apache-2.0
๐งญ The Zenyx Dataset Family: Which Dataset Should You Use?
Every Zenyx dataset lives under Arko007 and follows one convention: a messages column holding a JSON string of chat turns (roles system, user, assistant, tool), optional tools, and a src tag naming the family. The tables below route you to the right repo.
๐งฑ General SFT and training collections
| Dataset | Format | Volume | What it is | Best suited for |
|---|---|---|---|---|
Arko007/zenyx-sft-v2 |
Clean Parquet (train + validation) | 191,002 conversations, ~29.1M tokens | 45 domains of multi-turn, agentic, coding and reasoning chat; audited and de-duplicated | Standard supervised fine-tuning |
Arko007/Zenyx-100M-SFT-synth |
JSONL + Parquet archive | 209k accepted + 114k rejected rows, ~41M tokens | Raw generation traces and the log of every rejected row | Studying synthetic-data pipelines and failure modes |
Arko007/Zenyx-100M-SFT-data |
Pre-tokenised NumPy .npz shards (+ Parquet view) |
149,571 sequences of 2,049 tokens (208.2M trained tokens) | 15-source mixture, already tokenised, with loss masks | Zero-overhead GPU training of Zenyx-100M |
Arko007/zenyx-curated |
Parquet, 38 families | 2,520,243 rows | Filtered, de-duplicated, script-checked public and Zenyx-made sources | Building your own SFT mix family by family |
Arko007/zenyx-v3-synthetic-sft-v4 |
JSONL shards | 101,523 samples | Distilled reasoning, code and math dialogues from five public sources | Reasoning-heavy SFT |
Arko007/zenyx-v2-SFT-dataset |
Parquet, one config per source | about 187M rows (14 source sets) | Raw Nemotron SFT / RL, Cascade and math collections used for Zenyx V2 | Large-scale research on source data |
๐ฏ Instruction following, behaviour and constraints
| Dataset | Format | Volume | What it is | Best suited for |
|---|---|---|---|---|
Arko007/zenyx-instruct-exact (๐ You are here) |
Parquet, 33 configs | ~250M templated rows | Rule-based exact-instruction data; every answer computed by code | Exact counts, formats, letters, multi-question, follow-ups, rules that persist |
Arko007/zenyx-exact-mix |
Parquet, 5 group configs | 1,183,373 rows, ~180M tokens | Trainer-ready, de-duplicated subset of zenyx-instruct-exact | Dropping the exact-instruction data straight into a training mix |
Arko007/zenyx-constraint-write |
Parquet + JSONL | 394,998 rows, ~98.1M tokens | Exact-rule writing: word lengths, letter counts, lipograms, acrostics | Controllable, rule-exact text generation |
Arko007/zenyx-haiku-opus |
Parquet, 4 configs + raw JSONL | weak_games 2,496 / rest 14,843 / opus 40 / train 16,836 rows | Hand-written and solver-made behaviour data from Claude Haiku and Opus, in one repo | Calibration, tool recovery, games, multi-turn behaviour |
Arko007/zenyx-false-belief |
Parquet | 7,194 rows (v2) | Correcting false beliefs and false premises kindly, with true facts | Honesty and theory-of-mind behaviour |
Arko007/zenyx-frontier-handwritten |
JSONL, several rounds | about 15,000 multi-turn rows | Hand-written and repaired multi-turn dialogues in 55 reasoning domains | High-density multi-turn and tool arbitration supervision |
๐ ๏ธ Agents, tools and conversation
| Dataset | Format | Volume | What it is | Best suited for |
|---|---|---|---|---|
Arko007/zenyx-agentic-loops |
JSONL (accepted + rejects) | simulated-tool rollouts, counts in status/ |
Tool-use loops with executed simulated tools and verified rollouts | Agent-loop behaviour: errors, empty results, asking, confirming, polling |
Arko007/zenyx-spark-r11 |
Parquet, 8 families | 462,033 rows | Round-11 families: celebrity facts, dates, comprehension, slang, greetings, needle retrieval, STEM, code translation | Short, factual and conversational behaviour |
Arko007/zenyx-spark-r11-calib |
JSONL | 836 accepted rows | Small calibration run of the round-11 generators | Diagnostics only |
Arko007/zenyx-r11-extra |
Parquet, 3 folders | 95,334 rows | Math with executed code, casual chat and roleplay, rebuilt from three public sources | Tool-augmented math and conversational variety |
๐ข Reasoning, arithmetic and puzzles
| Dataset | Format | Volume | What it is | Best suited for |
|---|---|---|---|---|
Arko007/zenyx-arithmetic-100m |
Parquet, 21 shards | 100,000,000 rows | Exact integer arithmetic with verified step-by-step scratchpads | Teaching procedures instead of guessed results |
Arko007/zenyx-arc-synth |
Parquet (train + validation) | 597,098 + 2,000 tasks | Programmatically generated ARC-style grid puzzles, correct by construction | Spatial and inductive reasoning |
๐ฏ Quick navigation
- Standard fine-tuning on clean chat, agent and coding data โ
zenyx-sft-v2 - A big, mixable pool, chosen family by family โ
zenyx-curated - Exact counts, formats, letters, several questions in one prompt, one-word follow-ups, rules that must persist โ
zenyx-instruct-exact(everything) orzenyx-exact-mix(trainer-ready subset) - Rule-exact writing (letter counts, lipograms, acrostics) โ
zenyx-constraint-write - Calibration, tool recovery, games and multi-turn behaviour from Claude โ
zenyx-haiku-opus - Agent loops with executed tools โ
zenyx-agentic-loops - Arithmetic procedures at scale โ
zenyx-arithmetic-100m; grid reasoning โzenyx-arc-synth - Pre-tokenised shards for the Zenyx-100M trainer โ
Zenyx-100M-SFT-data
The models these datasets were built for: Arko007/Zenyx-100M-Base (Zenyx-Mini, about 151M parameters).
Dataset Overview
Rule-based synthetic chat data for exact instruction following, written by Python generators: no language model wrote any row. It targets tasks that ordinary instruction sets rarely contain: exact counts, exact formats, letter tasks, several questions or coding tasks in one prompt, follow-ups that refer back, rules that must hold for a whole conversation, and tool use with failures.
Most rows are short (about 100 to 500 tokens); persistent_rules rows are long (about 900 tokens, up to 35 turns). For training, use the de-duplicated, balanced subset in zenyx-exact-mix instead of the full set.
Categories (one config each)
| Config | Rows | What it contains |
|---|---|---|
exact_repeat |
10,000,000 | "Output exactly N of X": punctuation marks, letters, digits and words, with different separators |
length_limits |
3,907,649 | exactly N words, fewer than N words, exactly N sentences / bullets / lines |
alphabet_letters |
8,236,517 | counting a letter in a word, positions, next / previous / nth letter, the alphabet |
sequences |
10,000,000 | lists with an exact count (one per line), de-duplication, sums of listed numbers |
copy_transform |
4,452,235 | copy exactly, upper / lower case, nth word, first letters of each word |
format_only |
10,000,000 | answer in a fixed format only (yes / no, a number, a letter for a multiple-choice question) |
combos |
6,371,100 | several constraints at once (exact word count, case, ending, keyword, ordering, layout) |
contrast |
9,892,252 | plain questions that contain numbers or the word "exactly" and need a plain answer |
search_act |
10,000,000 | search with a tool, read the result, then finish the task (calculate, write a file, create an event, send an email, fetch a page); includes summary-only and not-found cases |
multi_question |
8,355,611 | several numbered questions in one prompt, answered in order (some include a code question) |
multi_code |
6,310,352 | two to four coding tasks in one prompt, every one answered with executed, tested code |
followups |
10,000,000 | multi-turn rows where later turns refer to earlier answers (double it, sort them, and Japan?) |
terse_followups |
10,000,000 | one-word follow-ups (ok, thanks, why, more, again, yes, no) answered briefly and ended |
token_surface |
10,000,000 | surface-form tasks: exact blank lines between words, snake_case, writing a number backwards or in words, exact character layouts |
trap_questions |
403,968 | classic trap word problems (the tempting answer is wrong) with the working shown |
pushback_holds |
10,000,000 | the user states a claim or pushes back; the answer holds to the computed truth (agrees when the user is right, disagrees when not) |
constants_decimals |
10,000,000 | decimal comparison, sorting decimals, simple constants |
tool_routing |
7,791,819 | decide between answering directly and calling a tool (calculator, search, time, weather); the tool result is then used |
long_exact |
10,975 | long exact-length outputs (a letter repeated many times, long exact lists) |
table_lookup |
10,000,000 | questions over a small markdown or CSV table (sort, highest, total, average) |
code_output |
2,046,699 | what does this Python snippet print (every answer comes from running the snippet) |
calendar |
10,000,000 | date arithmetic, weekday of a date, days between dates, day of the year |
word_problems |
8,721,494 | short arithmetic word problems (change, ratios, ages) with the working |
logic |
10,000,000 | ordering puzzles (faster / heavier / older than) and family-relation chains |
grid_moves |
10,000,000 | walk on a grid: final coordinates and facing direction after a list of moves and turns |
persistent_rules |
10,000,000 | a rule set in the first turn (capital letters only, one word, square brackets, start / end every reply with a phrase) that must hold for 10 to 35 later turns |
odd_count |
10,000,000 | odd-one-out and "how many of these are X" questions |
grammar_fix |
10,000,000 | fix grammar, punctuation and doubled words in a sentence |
text_extract |
7,092,187 | pull emails, dates, counts of a word or sentences out of a short text |
num_convert |
5,860,987 | rounding to the nearest ten / hundred and to a number of decimal places |
math_basic |
10,000,000 | arithmetic, brackets, mean, mode and similar |
bugfix_exec |
855 | a function with a bug and a failing example; the fix is executed against the expected result |
tool_recovery_gen |
348,145 | a tool call fails (error or empty result), the retry changes the query, and the answer says it came from the retry |
Rows are unique, but they are made from templates: ten million rows of one category are not ten million different ideas. Take a small, spread-out sample of each category rather than all of it. Four categories ran out of distinct rows and are small: long_exact, bugfix_exec, trap_questions, tool_recovery_gen.
How the rows are made
- Answers are computed by code: arithmetic, counting, sorting, dates and letter positions are calculated; code answers are executed and tested against a reference before they are written.
- Prompts come from phrase pools combined with many different values, so the same task appears in many wordings.
- Sentences come from a small grammar builder with agreement, articles and plurals handled by rule, and exact word counts by construction.
- Every row passes a lint for common grammar slips (a / an, "1 times", doubled words, empty slots); rows that fail are dropped.
- Each generation kernel keeps only rows whose hash falls in its own slice, so no row is written twice across the 32 generation runs.
Checks done after generation (20 Nov 2026)
A separate pass re-derived the answers with plain Python and compared them with the stored answers, on random shards from several generation kernels (about 130,000 rows in total):
- Re-computed and matching, no errors found (the checker understands the common prompt types of each category, a quarter to all of the sampled rows depending on the category; the rest were only structure-checked): code_output (10,000 of 10,000 snippets re-run), tool_routing (10,000 calculator calls), multi_code and bugfix_exec (code blocks run and their examples checked), calendar, alphabet_letters, exact_repeat, long_exact, num_convert, constants_decimals, math_basic, grid_moves, logic, word_problems, combos, length_limits, sequences, copy_transform, text_extract, table_lookup, token_surface, persistent_rules (the rule is checked on every later turn), multi_question (every question answered, code examples run), format_only, pushback_holds.
- Every equation written in the answers or the
<think>text (a + b x c = d) was re-evaluated in all categories: no wrong equation found. - Not independently re-computed (structure, lint and reading samples only): contrast, followups and terse_followups (their arithmetic was checked), grammar_fix, trap_questions, tool_recovery_gen, and the "how many" kind of odd_count (a few read by hand).
- Structure on every sampled row: valid JSON, roles in order, a final assistant turn, no empty turns, balanced
<think>tags, no control characters.
The reports are in qa/; per-kernel row counts are in manifests/.
Data format and schema
{
"messages": "[{\"role\": \"system\", \"content\": \"...\"}, {\"role\": \"user\", \"content\": \"...\"}, {\"role\": \"assistant\", \"content\": \"<think>...</think>\\nanswer\"}]",
"tools": "",
"src": "exact_repeat",
"meta": "{\"family\": \"...\", \"kind\": \"...\", \"kid\": 3}"
}
messages(string): JSON list of{role, content}turns. Assistant turns start with a short<think>plan, then the answer. Tool rows use<tool_call>[{"name": ..., "arguments": {...}}]</tool_call>in the assistant text andtoolturns holding JSON results.tools(string): empty (tool names are given in the system prompt text).src(string): the family tag;meta(string): JSON with family, kind and the generator id.
Usage
from datasets import load_dataset
import json
ds = load_dataset("Arko007/zenyx-instruct-exact", "followups", split="train", streaming=True)
row = next(iter(ds)); msgs = json.loads(row["messages"])
Limits
- Synthetic and templated: wording is varied but not natural writing. The
<think>text is templated and every row has one. - Some categories use only a handful of prompt templates (for example word_problems, calendar, logic, trap_questions): a model can learn the template instead of the skill if the share is large.
- The products, versions, events and prices in
search_actandtool_recovery_genare invented (example domains), not real facts. Country, capital, element and shape tables are short hand-written lists. - Generated sentences are grammatical by rule but can sound odd (for example, unusual adjective order). Grammar is checked by heuristics, not by a full grammar checker.
License and citation
This dataset is released under the Apache 2.0 License.
@dataset{zenyx_instruct_exact_2026,
author = {Arko and Zenyx AI Contributors},
title = {Zenyx Instruct Exact: Rule-Based Exact-Instruction Chat Data},
year = {2026},
publisher = {Hugging Face},
howpublished = {\url{https://huggingface.co/datasets/Arko007/zenyx-instruct-exact}}
}
- Downloads last month
- 14