Dataset Viewer
Auto-converted to Parquet Duplicate
id
string
prompt
string
text
string
axis
string
domain
string
family
string
group
string
source_identity
string
prompt_sha256
string
token_ids_sha256
string
language
string
topic
string
repo
string
file
string
article_path
string
problem_id
string
sample_id
string
schema_id
string
source
string
subset
string
source_url
string
source_revision
string
source_license
string
provenance_json
string
max_tokens
int32
prompt_tokens
int32
messages
list
cal-000
The function `getClassPathUrlsForNested` (lines 338-373) has been ablated from the file below. It is described as: This method contains the most complex logic for resolving nested archive paths, handling various URL formats, file existence checks, and archive instantiation. It is significantly larger and more intricate...
The function `getClassPathUrlsForNested` (lines 338-373) has been ablated from the file below. It is described as: This method contains the most complex logic for resolving nested archive paths, handling various URL formats, file existence checks, and archive instantiation. It is significantly larger and more intricate...
code_agentic
code
ablation
code:spring-projects/spring-boot
e71f50ef6d574e6b8e849d486c364deac9dd290354f85e92d0e383ce9fd181bf
62f0fe5da016fd25a923e1d810ad06a891261147f87326581f16da6dac1250ec
319ebab2f435c7159804f4980a7a3b4cc0396d4bd0d96e8a9a4594175c20555c
Java
null
spring-projects/spring-boot
loader/spring-boot-loader/src/main/java/org/springframework/boot/loader/launch/PropertiesLauncher.java
null
null
null
null
null
null
https://github.com/spring-projects/spring-boot/blob/ac2cfe03538b87a0dc9d2619ed52a4c901eaea54/loader/spring-boot-loader/src/main/java/org/springframework/boot/loader/launch/PropertiesLauncher.java
ac2cfe03538b87a0dc9d2619ed52a4c901eaea54
upstream-code-license
{"axis": "code_agentic", "domain": "code", "family": "ablation", "file": "loader/spring-boot-loader/src/main/java/org/springframework/boot/loader/launch/PropertiesLauncher.java", "group": "code:spring-projects/spring-boot", "id": "cal-000", "language": "Java", "prompt_sha256": "62f0fe5da016fd25a923e1d810ad06a891261147f...
8
768
[ { "role": "user", "content": "The function `getClassPathUrlsForNested` (lines 338-373) has been ablated from the file below. It is described as: This method contains the most complex logic for resolving nested archive paths, handling various URL formats, file existence checks, and archive instantiation. It ...
cal-001
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/enwiki/learning/columbia-gorge-community-college.md): # Columbia Gorge Community College 'Columbia Gorge Community College' (CGCC) is a public community college with campuses in The Dalles and Hood River, Or...
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/enwiki/learning/columbia-gorge-community-college.md): # Columbia Gorge Community College 'Columbia Gorge Community College' (CGCC) is a public community college with campuses in The Dalles and Hood River, Or...
general
wiki
mc
wiki:processed/wiki/enwiki/learning/columbia-gorge-community-college.md
d2eb8a0922a0154302312bf65a9e593bf0a875d65288be1a1e565d2c2c7b2305
d0db5cfc7c8780997a55d727a27cb4d57903c745518f07f5fc860476286a5618
48759fc1222d1678002afff36c27ed3fa464598fddd3d82f5256cba89985f26f
en
learning
null
null
processed/wiki/enwiki/learning/columbia-gorge-community-college.md
null
null
null
null
null
https://en.wikipedia.org/wiki/Columbia_Gorge_Community_College
20260701
cc-by-sa-4.0
{"article_path": "processed/wiki/enwiki/learning/columbia-gorge-community-college.md", "axis": "general", "domain": "wiki", "family": "mc", "group": "wiki:processed/wiki/enwiki/learning/columbia-gorge-community-college.md", "id": "cal-001", "language": "en", "prompt_sha256": "d0db5cfc7c8780997a55d727a27cb4d57903c745518...
8
767
[ { "role": "user", "content": "Use the reference article below to complete the task in the task's language.\n\nReference article (processed/wiki/enwiki/learning/columbia-gorge-community-college.md):\n# Columbia Gorge Community College\n\n'Columbia Gorge Community College' (CGCC) is a public community college...
cal-002
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/enwiki/popculture/atari.md): # Atari 'Atari' () is a brand name that has been owned by several entities since its inception in 1972. It is currently owned by French holding company Atari SA (formerly Infogra...
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/enwiki/popculture/atari.md): # Atari 'Atari' () is a brand name that has been owned by several entities since its inception in 1972. It is currently owned by French holding company Atari SA (formerly Infogra...
general
wiki
mc
wiki:processed/wiki/enwiki/popculture/atari.md
f610a9c8324cf75a46da78f7a6553ec57228a320c48d3c2c51380d8706129829
2f1f2ba48341b6ce6682be63b32713478615dddbbf806400b1bbe4c703462a08
868375af4a6a47d3edcf8df0f3977259a62fff9366f5527805df349bf2ea521b
en
popculture
null
null
processed/wiki/enwiki/popculture/atari.md
null
null
null
null
null
https://en.wikipedia.org/wiki/Atari
20260701
cc-by-sa-4.0
{"article_path": "processed/wiki/enwiki/popculture/atari.md", "axis": "general", "domain": "wiki", "family": "mc", "group": "wiki:processed/wiki/enwiki/popculture/atari.md", "id": "cal-002", "language": "en", "prompt_sha256": "2f1f2ba48341b6ce6682be63b32713478615dddbbf806400b1bbe4c703462a08", "prompt_tokens": 768, "sou...
8
768
[ { "role": "user", "content": "Use the reference article below to complete the task in the task's language.\n\nReference article (processed/wiki/enwiki/popculture/atari.md):\n# Atari\n\n'Atari' () is a brand name that has been owned by several entities since its inception in 1972. It is currently owned by Fr...
cal-003
题目:一瓶药有42片,张英每天吃3次,每次吃2片。这瓶药够吃多少天? 答案:7 题目:一群小朋友做游戏,游戏必须4人一组才能进行。现在一共有29个小朋友,最多能分成多少个小组做游戏? 答案:7 题目:一共有10个西瓜。有5只小猴来搬西瓜,每只小猴平均搬多少个? 答案:2 题目:小辉在超市买了4袋饼干付了8元,如果买8袋饼干要付多少元? 答案:16 题目:元旦到了,二(4)班买了42个气球装点教室。如果每6个气球扎成一束,可以扎几束? 答案:7 题目:三个小组一共收集了94个矿泉水瓶,第一组收集了34个,第二组收集了29个,第三组收集了多少个? 答案:31 题目:二年级一班有45人,每张桌子坐7人。要让每一位同学都有座位,...
题目:一瓶药有42片,张英每天吃3次,每次吃2片。这瓶药够吃多少天? 答案:7 题目:一群小朋友做游戏,游戏必须4人一组才能进行。现在一共有29个小朋友,最多能分成多少个小组做游戏? 答案:7 题目:一共有10个西瓜。有5只小猴来搬西瓜,每只小猴平均搬多少个? 答案:2 题目:小辉在超市买了4袋饼干付了8元,如果买8袋饼干要付多少元? 答案:16 题目:元旦到了,二(4)班买了42个气球装点教室。如果每6个气球扎成一束,可以扎几束? 答案:7 题目:三个小组一共收集了94个矿泉水瓶,第一组收集了34个,第二组收集了29个,第三组收集了多少个? 答案:31 题目:二年级一班有45人,每张桌子坐7人。要让每一位同学都有座位,...
math_reasoning
math
solve
math:d201c55bf847a9b8dc47620d7e4dcdb6be66f27899332facf2c85fa34b202a53
7e0892a68a5be564054fc959ffd6cab7b2f6d5ca6ba453d72c8e8a05f307a560
4bc1298bbddee10ae19778d00bfbd269da49114c00e6a3b2ddafdc1c9682729b
9e7486d6bf86a9e8aec4ec342ead7d30b178656ce92ae088eb616fc0a9b56a31
zh
math
null
null
null
d201c55bf847a9b8dc47620d7e4dcdb6be66f27899332facf2c85fa34b202a53
null
null
cmath
cmath
https://huggingface.co/datasets/weitianwen/cmath
792a5d5cc047e50d53780b5b38affb98a0a27a64
cc-by-4.0
{"axis": "math_reasoning", "domain": "math", "family": "solve", "group": "math:d201c55bf847a9b8dc47620d7e4dcdb6be66f27899332facf2c85fa34b202a53", "id": "cal-003", "language": "zh", "problem_id": "d201c55bf847a9b8dc47620d7e4dcdb6be66f27899332facf2c85fa34b202a53", "prompt_sha256": "4bc1298bbddee10ae19778d00bfbd269da49114...
8
1,019
[ { "role": "user", "content": "题目:一瓶药有42片,张英每天吃3次,每次吃2片。这瓶药够吃多少天?\n答案:7\n\n题目:一群小朋友做游戏,游戏必须4人一组才能进行。现在一共有29个小朋友,最多能分成多少个小组做游戏?\n答案:7\n\n题目:一共有10个西瓜。有5只小猴来搬西瓜,每只小猴平均搬多少个?\n答案:2\n\n题目:小辉在超市买了4袋饼干付了8元,如果买8袋饼干要付多少元?\n答案:16\n\n题目:元旦到了,二(4)班买了42个气球装点教室。如果每6个气球扎成一束,可以扎几束?\n答案:7\n\n题目:三个小组一共收集了94个矿泉水瓶,第一组收集了34个,第二组收...
cal-004
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/zhwiki/crypto/padding-cryptography.md): # 填充 (密码学) 在密码学中,'「填充」'指几种不同的操作。 ### 经典密码学 许多经典密码会将明文排列成特定的形状(如:正方形、长方形等),而如果明文不能完全符合形状,就需要添加字母来填满形状。 用无意义的字母来填充则更可以阻碍一些密码分析。 经典填充有时会造成误解,如有名的「the world wonders(全世界都...
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/zhwiki/crypto/padding-cryptography.md): # 填充 (密码学) 在密码学中,'「填充」'指几种不同的操作。 ### 经典密码学 许多经典密码会将明文排列成特定的形状(如:正方形、长方形等),而如果明文不能完全符合形状,就需要添加字母来填满形状。 用无意义的字母来填充则更可以阻碍一些密码分析。 经典填充有时会造成误解,如有名的「the world wonders(全世界都...
general
wiki
short
wiki:processed/wiki/zhwiki/crypto/padding-cryptography.md
8cdc5125f7546989ff1e57b042abf806da1676a6a5a7fc5f5ff8611338f88701
e9184f41121a200abb9534f6b94f9a619cf79f1a5e33a1ab139551a26b0a70f3
dcc5faa9cb4a702aaecc19b4980b89b152eafc599adf6a5ff635d3b0cd52f077
zh
crypto
null
null
processed/wiki/zhwiki/crypto/padding-cryptography.md
null
null
null
null
null
https://zh.wikipedia.org/wiki/%E5%A1%AB%E5%85%85_%28%E5%AF%86%E7%A0%81%E5%AD%A6%29
20260701
cc-by-sa-4.0
{"article_path": "processed/wiki/zhwiki/crypto/padding-cryptography.md", "axis": "general", "domain": "wiki", "family": "short", "group": "wiki:processed/wiki/zhwiki/crypto/padding-cryptography.md", "id": "cal-004", "language": "zh", "prompt_sha256": "e9184f41121a200abb9534f6b94f9a619cf79f1a5e33a1ab139551a26b0a70f3", "...
8
767
[ { "role": "user", "content": "Use the reference article below to complete the task in the task's language.\n\nReference article (processed/wiki/zhwiki/crypto/padding-cryptography.md):\n# 填充 (密码学)\n\n在密码学中,'「填充」'指几种不同的操作。\n\n### 经典密码学\n许多经典密码会将明文排列成特定的形状(如:正方形、长方形等),而如果明文不能完全符合形状,就需要添加字母来填满形状。 用无意义的字母来填充则更可以...
cal-005
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/enwiki/ai/artificial-intelligence.md): # Artificial intelligence 'Artificial intelligence' ('AI') is the capability of computational systems to perform tasks typically associated with human intelligence, suc...
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/enwiki/ai/artificial-intelligence.md): # Artificial intelligence 'Artificial intelligence' ('AI') is the capability of computational systems to perform tasks typically associated with human intelligence, suc...
reasoning_termination
wiki
mc
wiki:processed/wiki/enwiki/ai/artificial-intelligence.md
0d93e77a48c36b2c721ca2bc3fa6ff3fd8e45e567cfdc5b912a9df077d240337
b926ff2ec14e2c230d26bbe99ab8db4511ff83b9286f6c1391d3ce55c02a860b
70b274294cb1157dfc0b7bba13a8137044acec556d263b601d4ba6679fec986e
en
ai
null
null
processed/wiki/enwiki/ai/artificial-intelligence.md
null
null
null
null
null
https://en.wikipedia.org/wiki/Artificial_intelligence
20260701
cc-by-sa-4.0
{"article_path": "processed/wiki/enwiki/ai/artificial-intelligence.md", "axis": "reasoning_termination", "domain": "wiki", "family": "mc", "group": "wiki:processed/wiki/enwiki/ai/artificial-intelligence.md", "id": "cal-005", "language": "en", "prompt_sha256": "b926ff2ec14e2c230d26bbe99ab8db4511ff83b9286f6c1391d3ce55c02...
8
768
[ { "role": "user", "content": "Use the reference article below to complete the task in the task's language.\n\nReference article (processed/wiki/enwiki/ai/artificial-intelligence.md):\n# Artificial intelligence\n\n'Artificial intelligence' ('AI') is the capability of computational systems to perform tasks ty...
cal-006
You are improving the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. Output the ENTIRE improved file — complete and self-contained, with every original function preserved (names, signatures,...
You are improving the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. Output the ENTIRE improved file — complete and self-contained, with every original function preserved (names, signatures,...
code_agentic
code
rewrite
code:nodejs/node
19ba81509d6b2f96e5e654cd5659c95276d989ab7afb6bed537f57a94885d68e
e6ddafcd5c0d7d172730db59ddc2583641856204cb5cd49ef177fdc49959c269
b9da4268af4f8272d073a77b433195ad5223f9961f56df8a83f59bb53b5fa0b6
JavaScript
null
nodejs/node
lib/url.js
null
null
null
null
null
null
https://github.com/nodejs/node/blob/9e6bf8dbdeb890cbb09385b065f1c352128cd439/lib/url.js
9e6bf8dbdeb890cbb09385b065f1c352128cd439
upstream-code-license
{"axis": "code_agentic", "domain": "code", "family": "rewrite", "file": "lib/url.js", "group": "code:nodejs/node", "id": "cal-006", "language": "JavaScript", "prompt_sha256": "e6ddafcd5c0d7d172730db59ddc2583641856204cb5cd49ef177fdc49959c269", "prompt_tokens": 768, "repo": "nodejs/node", "source_identity": "19ba81509d6b...
8
768
[ { "role": "user", "content": "You are improving the file below.\n\nGoals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability.\n\nOutput the ENTIRE improved file — complete and self-contained, with every\n...
cal-007
Which isometry transformation correctly maps the line segment $\overline{CD}$ to $\overline{C'D'}$ if $C(3, -2)$ maps to $C'(-3, 2)$ and $D(2, -5)$ maps to $D'(-2, 5)$? - Reflection in the $y$-axis - Clockwise rotation around the origin by $180^{\circ}$ - Translation by 6 units to the left and 7 units up - Reflection i...
Which isometry transformation correctly maps the line segment $\overline{CD}$ to $\overline{C'D'}$ if $C(3, -2)$ maps to $C'(-3, 2)$ and $D(2, -5)$ maps to $D'(-2, 5)$? - Reflection in the $y$-axis - Clockwise rotation around the origin by $180^{\circ}$ - Translation by 6 units to the left and 7 units up - Reflection i...
math_reasoning
math
solve
math:08015d3a75e035eb21d17f819b4db36b89518e1618ad8d56efad08c2f4fc42c2
bfb6138b9f88e8d3145420286d994d519c70fb6de8bc6c2f1e5dd0ed6bc413a9
98392b84e71b9645120778b3288d4cebfde64c3e0d9c27da5c3d471cf36299ab
df63924c1d2125c51964b1f1a25ccec453aa865ece9e846f16c007660ae011aa
en
math
null
null
null
08015d3a75e035eb21d17f819b4db36b89518e1618ad8d56efad08c2f4fc42c2
null
null
numinamath-cot
synthetic_amc
https://huggingface.co/datasets/AI-MO/NuminaMath-CoT
792a5d5cc047e50d53780b5b38affb98a0a27a64
apache-2.0
{"axis": "math_reasoning", "domain": "math", "family": "solve", "group": "math:08015d3a75e035eb21d17f819b4db36b89518e1618ad8d56efad08c2f4fc42c2", "id": "cal-007", "language": "en", "problem_id": "08015d3a75e035eb21d17f819b4db36b89518e1618ad8d56efad08c2f4fc42c2", "prompt_sha256": "98392b84e71b9645120778b3288d4cebfde64c3...
8
720
[ { "role": "user", "content": "Which isometry transformation correctly maps the line segment $\\overline{CD}$ to $\\overline{C'D'}$ if $C(3, -2)$ maps to $C'(-3, 2)$ and $D(2, -5)$ maps to $D'(-2, 5)$?\n- Reflection in the $y$-axis\n- Clockwise rotation around the origin by $180^{\\circ}$\n- Translation by 6...
cal-008
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/zhwiki/physics/ding-zhaozhong.md): # 丁肇中 '丁肇中'(,),美國華裔實驗物理学家,生於美國密歇根州安阿伯。中央研究院、美国科学院院士及中國科學院外籍院士,現任美国麻省理工学院教授,曾获得1976年诺贝尔物理学奖。他知名於對J粒子的發現。 ### 生平 #### 入學之前 1936年,丁肇中出生在美国密歇根州的安阿伯,他的父母原本希望他出生在中國,不過當他父母在美國進行學...
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/zhwiki/physics/ding-zhaozhong.md): # 丁肇中 '丁肇中'(,),美國華裔實驗物理学家,生於美國密歇根州安阿伯。中央研究院、美国科学院院士及中國科學院外籍院士,現任美国麻省理工学院教授,曾获得1976年诺贝尔物理学奖。他知名於對J粒子的發現。 ### 生平 #### 入學之前 1936年,丁肇中出生在美国密歇根州的安阿伯,他的父母原本希望他出生在中國,不過當他父母在美國進行學...
reasoning_termination
wiki
relate
wiki:processed/wiki/zhwiki/physics/ding-zhaozhong.md
d532c3a2a0d69217580e350d129b5b140fab41780d50ec4ce82850202e1e4711
833b3b522483c256e7b90696214e28d946bf58df22cf0d321513757b95ece641
6e672947bfba70cba9cdf9ff19ca448eb30ea6003ba2a080b17a51c87c58bd1e
zh
physics
null
null
processed/wiki/zhwiki/physics/ding-zhaozhong.md
null
null
null
null
null
https://zh.wikipedia.org/wiki/%E4%B8%81%E8%82%87%E4%B8%AD
20260701
cc-by-sa-4.0
{"article_path": "processed/wiki/zhwiki/physics/ding-zhaozhong.md", "axis": "reasoning_termination", "domain": "wiki", "family": "relate", "group": "wiki:processed/wiki/zhwiki/physics/ding-zhaozhong.md", "id": "cal-008", "language": "zh", "prompt_sha256": "833b3b522483c256e7b90696214e28d946bf58df22cf0d321513757b95ece64...
8
768
[ { "role": "user", "content": "Use the reference article below to complete the task in the task's language.\n\nReference article (processed/wiki/zhwiki/physics/ding-zhaozhong.md):\n# 丁肇中\n\n'丁肇中'(,),美國華裔實驗物理学家,生於美國密歇根州安阿伯。中央研究院、美国科学院院士及中國科學院外籍院士,現任美国麻省理工学院教授,曾获得1976年诺贝尔物理学奖。他知名於對J粒子的發現。\n\n### 生平\n#### 入學之前\...
cal-009
Tool schemas: "[{\"name\": \"get_list_of_top_level_domains\", \"description\": \"Retrieve an exhaustive list of generic and country code top-level domains (TLDs).\", \"parameters\": {\"tlds\": {\"description\": \"A comma-separated string of specific TLDs to query. If not provided, all TLDs are retrieved.\", \"type\": \...
Tool schemas: "[{\"name\": \"get_list_of_top_level_domains\", \"description\": \"Retrieve an exhaustive list of generic and country code top-level domains (TLDs).\", \"parameters\": {\"tlds\": {\"description\": \"A comma-separated string of specific TLDs to query. If not provided, all TLDs are retrieved.\", \"type\": \...
structured_output
structured
tool_call
struct:42240
e52e9460682067d746cfc43260a12a57bfb8acb9f29c2c21cd31118d5f592d0b
9291db075f02bc203271a549ef643de251ab385a816ed44316abf652b3ea7c67
bbcd0b70551bc3466dda1b5feaf1c67c53dbd36061f84337513e0a3a8002d0d3
en
null
null
null
null
null
42240
07c7c8c2ec319b8920b06b8a0838c249e0e54689c91e3d29b4cccb649728ed5c
xlam-function-calling-60k
null
https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k
792a5d5cc047e50d53780b5b38affb98a0a27a64
cc-by-4.0
{"axis": "structured_output", "domain": "structured", "family": "tool_call", "group": "struct:42240", "id": "cal-009", "language": "en", "prompt_sha256": "9291db075f02bc203271a549ef643de251ab385a816ed44316abf652b3ea7c67", "prompt_tokens": 805, "sample_id": "42240", "schema_id": "07c7c8c2ec319b8920b06b8a0838c249e0e54689...
8
805
[ { "role": "user", "content": "Tool schemas:\n\"[{\\\"name\\\": \\\"get_list_of_top_level_domains\\\", \\\"description\\\": \\\"Retrieve an exhaustive list of generic and country code top-level domains (TLDs).\\\", \\\"parameters\\\": {\\\"tlds\\\": {\\\"description\\\": \\\"A comma-separated string of speci...
cal-010
Compute $$\sum_{k=1}^{1000} k(\lceil \log_{\sqrt{2}}{k}\rceil- \lfloor\log_{\sqrt{2}}{k} \rfloor).$$ To compute the sum $\sum_{k=1}^{1000} k(\lceil \log_{\sqrt{2}}{k}\rceil- \lfloor\log_{\sqrt{2}}{k} \rfloor)$, we first need to understand the behavior of the expression inside the sum, specifically $\lceil \log_{\sqrt{...
Compute $$\sum_{k=1}^{1000} k(\lceil \log_{\sqrt{2}}{k}\rceil- \lfloor\log_{\sqrt{2}}{k} \rfloor).$$ To compute the sum $\sum_{k=1}^{1000} k(\lceil \log_{\sqrt{2}}{k}\rceil- \lfloor\log_{\sqrt{2}}{k} \rfloor)$, we first need to understand the behavior of the expression inside the sum, specifically $\lceil \log_{\sqrt{...
math_reasoning
math
solve
math:1c7a2d4dd54bece15b1eae4f7d3d716a3f266e039404ca6686b7815e8a03914e
0ffbff986b36589b71698d99ebf34ca5772870585bedfc51b74f538820d2241f
662f5f984130b815f179ef8312d1392163c067dd49e272d921b247703a792760
bf91d08193275b2de90834065ed102ceefb608f2a334d89be315331085368834
en
math
null
null
null
1c7a2d4dd54bece15b1eae4f7d3d716a3f266e039404ca6686b7815e8a03914e
null
null
numinamath-cot
math
https://huggingface.co/datasets/AI-MO/NuminaMath-CoT
792a5d5cc047e50d53780b5b38affb98a0a27a64
apache-2.0
{"axis": "math_reasoning", "domain": "math", "family": "solve", "group": "math:1c7a2d4dd54bece15b1eae4f7d3d716a3f266e039404ca6686b7815e8a03914e", "id": "cal-010", "language": "en", "problem_id": "1c7a2d4dd54bece15b1eae4f7d3d716a3f266e039404ca6686b7815e8a03914e", "prompt_sha256": "662f5f984130b815f179ef8312d1392163c067d...
8
749
[ { "role": "user", "content": "Compute\n$$\\sum_{k=1}^{1000} k(\\lceil \\log_{\\sqrt{2}}{k}\\rceil- \\lfloor\\log_{\\sqrt{2}}{k} \\rfloor).$$\n\nTo compute the sum $\\sum_{k=1}^{1000} k(\\lceil \\log_{\\sqrt{2}}{k}\\rceil- \\lfloor\\log_{\\sqrt{2}}{k} \\rfloor)$, we first need to understand the behavior of t...
cal-011
You are performing a careful code review of the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. For each issue you find, provide: 1. The exact current snippet (with line numbers or context ma...
You are performing a careful code review of the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. For each issue you find, provide: 1. The exact current snippet (with line numbers or context ma...
code_agentic
code
review
code:spring-projects/spring-boot
d481f3947ce34863310ae572b8efe266587683efcac5346bfd3e695438149e68
145f574448a18bf6ead9e340bbe6343a584f3c4ae165d85e6c6cf1d4f21d1616
a8f11006a5b67cd70f972fa2f198748120a246874a29a44b0bcaf58f80ec5ddb
Java
null
spring-projects/spring-boot
core/spring-boot-autoconfigure/src/main/java/org/springframework/boot/autoconfigure/condition/ConditionEvaluationReport.java
null
null
null
null
null
null
https://github.com/spring-projects/spring-boot/blob/ac2cfe03538b87a0dc9d2619ed52a4c901eaea54/core/spring-boot-autoconfigure/src/main/java/org/springframework/boot/autoconfigure/condition/ConditionEvaluationReport.java
ac2cfe03538b87a0dc9d2619ed52a4c901eaea54
upstream-code-license
{"axis": "code_agentic", "domain": "code", "family": "review", "file": "core/spring-boot-autoconfigure/src/main/java/org/springframework/boot/autoconfigure/condition/ConditionEvaluationReport.java", "group": "code:spring-projects/spring-boot", "id": "cal-011", "language": "Java", "prompt_sha256": "145f574448a18bf6ead9e...
8
768
[ { "role": "user", "content": "You are performing a careful code review of the file below.\n\nGoals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability.\n\nFor each issue you find, provide:\n1. The exact c...
cal-012
You are improving the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. Output the ENTIRE improved file — complete and self-contained, with every original function preserved (names, signatures,...
You are improving the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. Output the ENTIRE improved file — complete and self-contained, with every original function preserved (names, signatures,...
code_agentic
code
rewrite
code:TileLang/tvm
bba5100e6135c0b2688b3e4c5bed98c50fe64686523474c61cbe2ee0f57204c8
973c48d73105b4449009a5101146068103254686a57ee55051f6373e30f4216f
382595f343dc834256f0acaf603ee4fde64abb6b12f7822a60a40e7a3ac5f013
TVM
null
TileLang/tvm
python/tvm/tirx/build.py
null
null
null
null
null
null
https://github.com/TileLang/tvm/blob/3a10c4d8551b6c23d7b0b843b1fc9ef78af320c0/python/tvm/tirx/build.py
3a10c4d8551b6c23d7b0b843b1fc9ef78af320c0
upstream-code-license
{"axis": "code_agentic", "domain": "code", "family": "rewrite", "file": "python/tvm/tirx/build.py", "group": "code:TileLang/tvm", "id": "cal-012", "language": "TVM", "prompt_sha256": "973c48d73105b4449009a5101146068103254686a57ee55051f6373e30f4216f", "prompt_tokens": 768, "repo": "TileLang/tvm", "source_identity": "bba...
8
768
[ { "role": "user", "content": "You are improving the file below.\n\nGoals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability.\n\nOutput the ENTIRE improved file — complete and self-contained, with every\n...
cal-013
You are performing a careful code review of the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. For each issue you find, provide: 1. The exact current snippet (with line numbers or context ma...
You are performing a careful code review of the file below. Goals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability. For each issue you find, provide: 1. The exact current snippet (with line numbers or context ma...
code_agentic
code
review
code:microsoft/vscode
460e5b681f33232e32065851d34e1f8d7a36861be283797589319f33eb98568e
c428f9507746864749f4c4bbb4df0dae76eebb6b0c6b7dd8234a9b5f106fdda5
7365b9f24516b52ec58b5cb04fbfba48efdb75b3efb8cbe9d9b78873003a2663
TypeScript
null
microsoft/vscode
src/vs/base/common/cancellation.ts
null
null
null
null
null
null
https://github.com/microsoft/vscode/blob/3a03d6f72d628a7741c29f456b4ddbb5ae68502c/src/vs/base/common/cancellation.ts
3a03d6f72d628a7741c29f456b4ddbb5ae68502c
upstream-code-license
{"axis": "code_agentic", "domain": "code", "family": "review", "file": "src/vs/base/common/cancellation.ts", "group": "code:microsoft/vscode", "id": "cal-013", "language": "TypeScript", "prompt_sha256": "c428f9507746864749f4c4bbb4df0dae76eebb6b0c6b7dd8234a9b5f106fdda5", "prompt_tokens": 768, "repo": "microsoft/vscode",...
8
768
[ { "role": "user", "content": "You are performing a careful code review of the file below.\n\nGoals: improve naming and clarity, optimize performance where it matters, fix any latent bugs, harden error handling, and improve readability and maintainability.\n\nFor each issue you find, provide:\n1. The exact c...
cal-014
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/ptwiki/psych/condicionamento-classico.md): # Condicionamento clássico O 'condicionamento clássico' (ou 'condicionamento pavloviano' 'respondente') é um processo que descreve a gênese e a modificação de algun...
Use the reference article below to complete the task in the task's language. Reference article (processed/wiki/ptwiki/psych/condicionamento-classico.md): # Condicionamento clássico O 'condicionamento clássico' (ou 'condicionamento pavloviano' 'respondente') é um processo que descreve a gênese e a modificação de algun...
general
wiki
relate
wiki:processed/wiki/ptwiki/psych/condicionamento-classico.md
59de943c731c98aad1e8a8adef9a75da03a371a9e2ecf5957a6fa450afa7c05a
9e3042775333905b6a735619887a5cb1be78ff6065f7b08511e15db7703916cc
c4cfff4bda93af243ff935096d2fbf69969bf37f815e9ba2a1614e91ac2d95a0
pt
psych
null
null
processed/wiki/ptwiki/psych/condicionamento-classico.md
null
null
null
null
null
https://pt.wikipedia.org/wiki/Condicionamento_cl%C3%A1ssico
20260701
cc-by-sa-4.0
{"article_path": "processed/wiki/ptwiki/psych/condicionamento-classico.md", "axis": "general", "domain": "wiki", "family": "relate", "group": "wiki:processed/wiki/ptwiki/psych/condicionamento-classico.md", "id": "cal-014", "language": "pt", "prompt_sha256": "9e3042775333905b6a735619887a5cb1be78ff6065f7b08511e15db770391...
8
767
[ { "role": "user", "content": "Use the reference article below to complete the task in the task's language.\n\nReference article (processed/wiki/ptwiki/psych/condicionamento-classico.md):\n# Condicionamento clássico\n\nO 'condicionamento clássico' (ou 'condicionamento pavloviano' 'respondente') é um processo...
cal-015
The function `bats_encode_test_name` (lines 4-33) has been ablated from the file below. It is described as: This function contains the most complex logic in the file, involving string manipulation, regex matching, and character-by-character iteration to encode test names. It is approximately 30 lines long, which exceed...
The function `bats_encode_test_name` (lines 4-33) has been ablated from the file below. It is described as: This function contains the most complex logic in the file, involving string manipulation, regex matching, and character-by-character iteration to encode test names. It is approximately 30 lines long, which exceed...
code_agentic
code
ablation
code:bats-core/bats-core
ddc281810a8e31d4d15b5762c45274ba7c588bff1fec6c53aa38e293292a3367
a2ed2075ee9a3adb9f1b61d3ab62d218bc695b9bc327481f3563d3e0621c650d
b7a505980ce35b9f429c61fae5032e1b3503935b5878d924c037abb412578a0a
Shell
null
bats-core/bats-core
libexec/bats-core/bats-preprocess
null
null
null
null
null
null
https://github.com/bats-core/bats-core/blob/eb7f42f8d608ac693d7a4b67474f6714ea68cfc5/libexec/bats-core/bats-preprocess
eb7f42f8d608ac693d7a4b67474f6714ea68cfc5
upstream-code-license
{"axis": "code_agentic", "domain": "code", "family": "ablation", "file": "libexec/bats-core/bats-preprocess", "group": "code:bats-core/bats-core", "id": "cal-015", "language": "Shell", "prompt_sha256": "a2ed2075ee9a3adb9f1b61d3ab62d218bc695b9bc327481f3563d3e0621c650d", "prompt_tokens": 767, "repo": "bats-core/bats-core...
8
767
[ { "role": "user", "content": "The function `bats_encode_test_name` (lines 4-33) has been ablated from the file below. It is described as: This function contains the most complex logic in the file, involving string manipulation, regex matching, and character-by-character iteration to encode test names. It is...
cal-016
The least common multiple of $a$ and $b$ is $12$, and the least common multiple of $b$ and $c$ is $15$. What is the least possible value of the least common multiple of $a$ and $c$? $\textbf{(A) }20\qquad\textbf{(B) }30\qquad\textbf{(C) }60\qquad\textbf{(D) }120\qquad \textbf{(E) }180$ 1. **Identify the constraints gi...
The least common multiple of $a$ and $b$ is $12$, and the least common multiple of $b$ and $c$ is $15$. What is the least possible value of the least common multiple of $a$ and $c$? $\textbf{(A) }20\qquad\textbf{(B) }30\qquad\textbf{(C) }60\qquad\textbf{(D) }120\qquad \textbf{(E) }180$ 1. **Identify the constraints gi...
math_reasoning
math
solve
math:138eb9b54853c81575c2e1bb9e3968c4ccd70141e3d22bb3acf2ed39a1276611
8b7f2bf3c19b0d3e4f8ae49c06d6c89d13223cc2bff7cb81980bdfcfb1a440eb
ac4968b7d73401d6d70c93b4ccd9791aa7796f76a66e18d00af8778095d7e66d
b81d1bc7fd0e0a52c9067716882fa7d292c27cc4bb01c31761dfeafba1dc9114
en
math
null
null
null
138eb9b54853c81575c2e1bb9e3968c4ccd70141e3d22bb3acf2ed39a1276611
null
null
numinamath-cot
amc_aime
https://huggingface.co/datasets/AI-MO/NuminaMath-CoT
792a5d5cc047e50d53780b5b38affb98a0a27a64
apache-2.0
{"axis": "math_reasoning", "domain": "math", "family": "solve", "group": "math:138eb9b54853c81575c2e1bb9e3968c4ccd70141e3d22bb3acf2ed39a1276611", "id": "cal-016", "language": "en", "problem_id": "138eb9b54853c81575c2e1bb9e3968c4ccd70141e3d22bb3acf2ed39a1276611", "prompt_sha256": "ac4968b7d73401d6d70c93b4ccd9791aa7796f7...
8
556
[ { "role": "user", "content": "The least common multiple of $a$ and $b$ is $12$, and the least common multiple of $b$ and $c$ is $15$. What is the least possible value of the least common multiple of $a$ and $c$?\n$\\textbf{(A) }20\\qquad\\textbf{(B) }30\\qquad\\textbf{(C) }60\\qquad\\textbf{(D) }120\\qquad ...
End of preview. Expand in Data Studio

Agentic Quant Panel v2

A calibration panel for post-training quantization. It has two components that are used together:

Component What it is How a model consumes it
agentic Agentic Quant Panel v1, byte for byte: 1,426 prompts (about 1.08M tokens) of code, multilingual reference text, worked mathematics and function calling, with calibration, screening and heldout splits One user message per prompt, rendered with the target model's chat template, one unpadded prompt per batch
exl3_standard The ExLlamaV3 standard calibration data (C4, code, multilingual, technical, Wikipedia, tiny stories, plus random-token rows), pre-split into 2,726 ordered text segments Raw text, packed into fixed-width rows with the target tokenizer: 250 × 2,048 tokens (about 512K) by default

The calibration stream is agentic (dataset order), then exl3_standard (packed rows in contract order), about 1.59M tokens in total. exl3_standard is calibration-only, with no screening or heldout split.

Neither component contains model outputs or generated reasoning traces. Every token is a prompt or raw source text.

Why combine them

The two components cover different ground. agentic reproduces the workload of an MoE model used for coding, reasoning and tool use, through the chat template, in the model's own prompt format. exl3_standard is the corpus that ExLlamaV3 quantizes with by default: broad raw web, encyclopedic, narrative, source-code and technical text, plus random token rows that touch the whole vocabulary. The random rows exist to put calibration signal on embeddings and experts that natural text rarely reaches. With more distinct tokens and contexts, more routed experts receive natural observations, so fewer need near-expert or identity recovery.

Using exl3_standard

exl3_standard/segments.jsonl (and data/exl3_standard.parquet) hold the text units ExLlamaV3's builder feeds to its packer, already split and ordered:

Source Weight Upstream processor Segments Unit
c4 20 shuffle_lines (lines, random.seed(0) shuffle) 585 article
code 20 split_raw 1 stream
multilingual 10 shuffle_lines 728 article
technical 10 split_raw 1 stream
wiki 50 split_wiki (one article per <doc>, body only, more than 50 chars) 1,094 article
tiny 5 split_tiny (one story per <|endoftext|>) 317 article
random 20 random_data (torch.manual_seed(0), uniform over the vocabulary) — generated

Rows per source are max(1, int(weight / 135 × cal_rows)); the random source fills the remainder. At 250 rows that's 37/37/18/18/92/9 + 39 random.

Packing depends on the target tokenizer, so it isn't precomputed. Use the bundled packer, which implements contract exl3-standard-pack-v1:

pip install torch tokenizers safetensors
python reproduction/pack_exl3_standard.py . /path/to/model ./exl3_standard.safetensors --cal-rows 250 --cal-cols 2048

It writes input_ids [250, 2048] int64 plus a JSON receipt with each row's source, segment span, BOS/EOS flag and token-ID hash. Tokenization goes through ExLlamaV3's own Tokenizer class, vendored unchanged from the pinned upstream commit (exl3_standard/upstream/tokenizer.py). So BOS/EOS resolution, unspecial added tokens and the vocabulary size match upstream.

Exactness. The packed rows are bit-identical to upstream get_default_calibration() at commit 8c593e4e5c88da76a4aaf8f1b1838bb7c1fce4e0, checked on the tencent/Hy4-flash and deepseek-ai/DeepSeek-V4-Flash-0731 tokenizers at 250 × 2,048.

Rules.

  • Don't apply a chat template to these rows.
  • Don't use random rows in held-out metrics, because they aren't language.
  • Article rows alternate: even rows encode each article with BOS and EOS, odd rows without. A row takes whole articles until it reaches cal_cols, and the remainder of the cut article is dropped.

Using agentic

See v1/README.md, the unchanged v1 card, for its design, splits, schema, provenance and reproduction. Its files keep their v1 paths and hashes: calibration.jsonl, screening.jsonl, heldout.jsonl, manifest.json, statistics.json, publication.json, data/*.parquet, provenance/, reproduction/ and licenses/. Loaders that validated v1 by file hash validate these files unchanged.

Manifests and integrity

  • panel.json: both components, their files and how each is consumed.
  • exl3_standard/manifest.json: upstream repository, commit and builder hash; each source file's SHA-256, weight and processor; the segment file's SHA-256; and the packing contract.
  • exl3_standard/source/*.utf8: the original upstream files, byte for byte.
  • SHA256SUMS: every file in the release.

Each segment carries text_sha256, and the packer refuses a segment whose hash differs.

Licensing and attribution

This is a mixed-source compilation, and each component keeps its upstream terms.

  • agentic: see the v1 licensing section.
  • exl3_standard: the calibration data and code are from ExLlamaV3 by turboderp, MIT License (exl3_standard/upstream/LICENSE). The underlying texts (C4 web text, Wikipedia, TinyStories, public source code and technical documents) keep their original sources' terms. Wikipedia text is CC BY-SA 4.0.

Thanks to turboderp for the ExLlamaV3 calibration set, and to the v1 source authors.

@misc{agentic_quant_panel_v2,
  author = {wrldsuksgo2mars},
  title = {Agentic Quant Panel v2},
  year = {2026},
  howpublished = {Hugging Face dataset},
  url = {https://huggingface.co/datasets/wrldsuksgo2mars/agentic-quant-panel-v2}
}
Downloads last month
-