nlile commited on
Commit
8358c4d
·
verified ·
1 Parent(s): 0ef0b45

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +214 -0
README.md CHANGED
@@ -1,4 +1,17 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  dataset_info:
3
  features:
4
  - name: problem
@@ -31,3 +44,204 @@ configs:
31
  - split: train
32
  path: data/train-*
33
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: apache-2.0
3
+ task_categories:
4
+ - text-generation
5
+ - question-answering
6
+ language:
7
+ - en
8
+ tags:
9
+ - math
10
+ - post-training
11
+ - RL
12
+ - verifiable
13
+ - reasoning
14
+ pretty_name: NuminaMath 1.5 RL Verifiable
15
  dataset_info:
16
  features:
17
  - name: problem
 
44
  - split: train
45
  path: data/train-*
46
  ---
47
+
48
+ # Dataset Card for NuminaMath-1.5-RL-Verifiable
49
+
50
+ ## Dataset Description
51
+
52
+ - **Homepage:** https://huggingface.co/datasets/nlile/NuminaMath-1.5-RL-Verifiable
53
+ - **Repository:** [NuminaMath-1.5-RL-Verifiable](https://huggingface.co/datasets/nlile/NuminaMath-1.5-RL-Verifiable)
54
+ - **Based on:** [NuminaMath-1.5](https://huggingface.co/datasets/AI-MO/NuminaMath-1.5)
55
+
56
+ ### Dataset Summary
57
+
58
+ NuminaMath-1.5-RL-Verifiable is a curated subset of the NuminaMath-1.5 dataset, specifically filtered to support reinforcement learning applications requiring verifiable outcomes. This collection consists of 131,063 math word problems from the original dataset that meet strict filtering criteria: all problems have definitive numerical answers, validated problem statements and solutions, and come from high-quality, non-synthetic sources.
59
+
60
+ The filtering process removes multiple-choice questions, proofs, problems without clear numerical answers, and all synthetic content, while preserving the rich diversity of mathematical domains from the original collection.
61
+
62
+ ### Filtering Methodology
63
+
64
+ The dataset was created by applying the following filters to the original NuminaMath-1.5 dataset:
65
+
66
+ - **Removed question types**: Multiple-choice questions and proofs
67
+ - **Answer validation**: Retained only problems with non-empty, numerical answers (excluded 'proof', 'notfound' answers)
68
+ - **Source selection**: Excluded potentially lower-quality sources (cn_k12, orca_math, synthetic_math, metamath)
69
+ - **Quality filters**: Retained only problems with validated problem statements and solutions
70
+ - **Authenticity**: Excluded all synthetic problems
71
+
72
+ These filtering steps reduced the original dataset from 896,215 problems to 131,063 problems (approximately 14.6% of the original dataset), all with verifiable outcomes.
73
+
74
+ ## Dataset Structure
75
+
76
+ ### Data Instances
77
+
78
+ Each instance in the dataset contains:
79
+ - A math word problem statement
80
+ - A Chain of Thought (CoT) solution
81
+ - A definitive numerical answer
82
+ - Problem metadata including math domain type
83
+
84
+ ### Data Fields
85
+
86
+ - `problem`: Text description of the mathematical problem
87
+ - `ref_solution`: Step-by-step Chain of Thought (CoT) solution
88
+ - `answer`: Definitive numerical result
89
+ - `problem_type`: Mathematical domain (Algebra, Geometry, Number Theory, etc.)
90
+ - `question_type`: Always "math-word-problem" in this filtered dataset
91
+ - `source`: Origin of the problem (olympiads, cn_contest, aops_forum, etc.)
92
+ - `problem_is_valid`: Always "Yes" in this filtered dataset
93
+ - `solution_is_valid`: Always "Yes" in this filtered dataset
94
+ - `synthetic`: Always false in this filtered dataset
95
+
96
+ ### Dataset Statistics
97
+
98
+ #### Distribution by Source
99
+
100
+ | Source | Problem Count |
101
+ |--------|---------------|
102
+ | olympiads | 92,487 |
103
+ | cn_contest | 15,828 |
104
+ | aops_forum | 15,092 |
105
+ | amc_aime | 4,893 |
106
+ | inequalities | 1,145 |
107
+ | olympiads_ref | 1,001 |
108
+ | number_theory | 617 |
109
+ | **Total** | **131,063** |
110
+
111
+ #### Distribution by Problem Type
112
+
113
+ | Problem Type | Problem Count | Percentage |
114
+ |--------------|---------------|------------|
115
+ | Algebra | 42,972 | 32.79% |
116
+ | Geometry | 31,405 | 23.96% |
117
+ | Number Theory | 22,071 | 16.84% |
118
+ | Combinatorics | 17,144 | 13.08% |
119
+ | Logic and Puzzles | 7,250 | 5.53% |
120
+ | Calculus | 4,954 | 3.78% |
121
+ | Inequalities | 4,000 | 3.05% |
122
+ | Other | 1,267 | 0.97% |
123
+
124
+ #### Detailed Breakdown by Problem Type and Source
125
+
126
+ <details>
127
+ <summary>Click to expand detailed breakdown</summary>
128
+
129
+ **Algebra**
130
+ - olympiads: 31,752
131
+ - cn_contest: 6,776
132
+ - amc_aime: 1,886
133
+ - aops_forum: 1,684
134
+ - inequalities: 531
135
+ - olympiads_ref: 265
136
+ - number_theory: 78
137
+
138
+ **Geometry**
139
+ - olympiads: 22,091
140
+ - cn_contest: 4,377
141
+ - aops_forum: 3,316
142
+ - amc_aime: 1,454
143
+ - olympiads_ref: 99
144
+ - inequalities: 60
145
+ - number_theory: 8
146
+
147
+ **Number Theory**
148
+ - olympiads: 14,848
149
+ - aops_forum: 3,614
150
+ - cn_contest: 1,916
151
+ - amc_aime: 744
152
+ - number_theory: 489
153
+ - olympiads_ref: 329
154
+ - inequalities: 131
155
+
156
+ **Combinatorics**
157
+ - olympiads: 11,219
158
+ - aops_forum: 3,176
159
+ - cn_contest: 1,724
160
+ - amc_aime: 612
161
+ - olympiads_ref: 266
162
+ - inequalities: 125
163
+ - number_theory: 22
164
+
165
+ **Logic and Puzzles**
166
+ - olympiads: 5,677
167
+ - aops_forum: 1,197
168
+ - cn_contest: 212
169
+ - amc_aime: 136
170
+ - inequalities: 16
171
+ - number_theory: 7
172
+ - olympiads_ref: 5
173
+
174
+ **Calculus**
175
+ - olympiads: 3,894
176
+ - aops_forum: 907
177
+ - cn_contest: 139
178
+ - inequalities: 8
179
+ - amc_aime: 4
180
+ - olympiads_ref: 1
181
+ - number_theory: 1
182
+
183
+ **Inequalities**
184
+ - olympiads: 2,292
185
+ - aops_forum: 717
186
+ - cn_contest: 657
187
+ - inequalities: 273
188
+ - olympiads_ref: 34
189
+ - amc_aime: 25
190
+ - number_theory: 2
191
+
192
+ **Other**
193
+ - olympiads: 714
194
+ - aops_forum: 481
195
+ - amc_aime: 32
196
+ - cn_contest: 27
197
+ - number_theory: 10
198
+ - olympiads_ref: 2
199
+ - inequalities: 1
200
+
201
+ </details>
202
+
203
+ #### Original NuminaMath-1.5 Source Breakdown
204
+
205
+ | source | problems | question_type:proof | question_type:mcq | question_type:word |
206
+ |:---------------|-----------:|----------------------:|--------------------:|---------------------:|
207
+ | olympiads | 197084 | 62970 | 13529 | 117845 |
208
+ | olympiads_ref | 3638 | 2246 | nan | 1392 |
209
+ | amc_aime | 5872 | 208 | 4374 | 963 |
210
+ | aops_forum | 67841 | 24532 | 5924 | 33486 |
211
+ | cn_contest | 29944 | 8663 | 5602 | 15649 |
212
+ | inequalities | 7314 | 5780 | 49 | 1478 |
213
+ | number_theory | 4043 | 2591 | 15 | 1239 |
214
+ | cn_k12 | 268819 | 3966 | 115800 | 149010 |
215
+ | orca_math | 151934 | 1 | 17 | 151916 |
216
+ | synthetic_math | 148712 | 41 | 1057 | 147612 |
217
+ | metamath | 11014 | nan | 82 | 10932 |
218
+ | Total | 896215 | 110998 | 146449 | 631522 |
219
+
220
+
221
+ ## Additional Information
222
+
223
+ ### Licensing Information
224
+
225
+ The dataset follows the licensing of the original NuminaMath-1.5 dataset and is available under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0).
226
+
227
+ ### Citation Information
228
+
229
+ ```
230
+ @misc{nlile2025numinamath15rlverifiable,
231
+ author = {nlile},
232
+ title = {NuminaMath-1.5-RL-Verifiable},
233
+ year = {2025},
234
+ publisher = {Hugging Face},
235
+ journal = {Hugging Face Dataset Repository},
236
+ howpublished = {\url{https://huggingface.co/datasets/nlile/NuminaMath-1.5-RL-Verifiable}}
237
+ }
238
+
239
+ @misc{numina_math_datasets,
240
+ author = {Jia LI and Edward Beeching and Lewis Tunstall and Ben Lipkin and Roman Soletskyi and Shengyi Costa Huang and Kashif Rasul and Longhui Yu and Albert Jiang and Ziju Shen and Zihan Qin and Bin Dong and Li Zhou and Yann Fleureau and Guillaume Lample and Stanislas Polu},
241
+ title = {NuminaMath},
242
+ year = {2024},
243
+ publisher = {Numina},
244
+ journal = {Hugging Face repository},
245
+ howpublished = {\url{https://github.com/project-numina/aimo-progress-prize/blob/main/report/numina_dataset.pdf}}
246
+ }
247
+ ```