EtashGuha commited on
Commit
7779eba
·
verified ·
1 Parent(s): 137b435

Upload configuration file

Browse files
Files changed (1) hide show
  1. config/herorun3_math.yaml +228 -0
config/herorun3_math.yaml ADDED
@@ -0,0 +1,228 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ operators:
2
+ - id: seed_math_automath_1000000
3
+ config:
4
+ type: load_preexisting
5
+ framework_name: seed_math_automath_1000000
6
+ - id: drop_columns_automath_1000000
7
+ config:
8
+ type: function
9
+ function: data_strategies.commons.keep_only_columns
10
+ function_config:
11
+ columns_to_keep:
12
+ - instruction_seed
13
+ input_ids:
14
+ - seed_math_automath_1000000
15
+ - id: add_source_automath
16
+ config:
17
+ type: function
18
+ function: data_strategies.commons.add_constant_columns
19
+ function_config:
20
+ source: "automath"
21
+ input_ids:
22
+ - drop_columns_automath_1000000
23
+ - id: classify_automath
24
+ config:
25
+ type: fasttext
26
+ num_cpus: 32
27
+ hf_repo_id: mlfoundations-dev/instruction_filtering_fast_text_classifier_seed_math_best_s1
28
+ input_column: instruction_seed
29
+ target_label: "__label__QA_doc"
30
+ top_percentage_ranking: 10
31
+ input_ids:
32
+ - add_source_automath
33
+ - id: sample_dataset_automath
34
+ config:
35
+ type: function
36
+ function: data_strategies.commons.uniform_sample_fixed
37
+ function_config:
38
+ num_samples: 15_000
39
+ input_ids:
40
+ - classify_automath
41
+
42
+
43
+
44
+ - id: seed_math_nvidia_math_exploit
45
+ config:
46
+ type: load_preexisting
47
+ framework_name: seed_math_nvidia_math_exploit
48
+ - id: drop_columns_nvidia_math_exploit
49
+ config:
50
+ type: function
51
+ function: data_strategies.commons.keep_only_columns
52
+ function_config:
53
+ columns_to_keep:
54
+ - instruction_seed
55
+ input_ids:
56
+ - seed_math_nvidia_math_exploit
57
+ - id: add_source_nvidia_math
58
+ config:
59
+ type: function
60
+ function: data_strategies.commons.add_constant_columns
61
+ function_config:
62
+ source: "nvidia_math"
63
+ input_ids:
64
+ - drop_columns_nvidia_math_exploit
65
+ - id: classify_nvidia_math
66
+ config:
67
+ type: fasttext
68
+ num_cpus: 32
69
+ hf_repo_id: mlfoundations-dev/instruction_filtering_fast_text_classifier_seed_math_best_s1
70
+ input_column: instruction_seed
71
+ target_label: "__label__QA_doc"
72
+ top_percentage_ranking: 10
73
+ input_ids:
74
+ - add_source_nvidia_math
75
+ - id: sample_dataset_nvidia_math
76
+ config:
77
+ type: function
78
+ function: data_strategies.commons.uniform_sample_fixed
79
+ function_config:
80
+ num_samples: 15_000
81
+ input_ids:
82
+ - classify_nvidia_math
83
+
84
+
85
+ - id: seed_math_tiger_lab_math_exploit
86
+ config:
87
+ type: load_preexisting
88
+ framework_name: seed_math_tiger_lab_math_exploit
89
+ - id: drop_columns_tiger_lab_math_exploit
90
+ config:
91
+ type: function
92
+ function: data_strategies.commons.keep_only_columns
93
+ function_config:
94
+ columns_to_keep:
95
+ - instruction_seed
96
+ input_ids:
97
+ - seed_math_tiger_lab_math_exploit
98
+ - id: add_source_tiger_lab_math
99
+ config:
100
+ type: function
101
+ function: data_strategies.commons.add_constant_columns
102
+ function_config:
103
+ source: "tiger_lab_math"
104
+ input_ids:
105
+ - drop_columns_tiger_lab_math_exploit
106
+ - id: classify_tiger_lab_math
107
+ config:
108
+ type: fasttext
109
+ num_cpus: 32
110
+ hf_repo_id: mlfoundations-dev/instruction_filtering_fast_text_classifier_seed_math_best_s1
111
+ input_column: instruction_seed
112
+ target_label: "__label__QA_doc"
113
+ top_percentage_ranking: 10
114
+ input_ids:
115
+ - add_source_tiger_lab_math
116
+ - id: sample_dataset_tiger_lab_math
117
+ config:
118
+ type: function
119
+ function: data_strategies.commons.uniform_sample_fixed
120
+ function_config:
121
+ num_samples: 15_000
122
+ input_ids:
123
+ - classify_tiger_lab_math
124
+
125
+ - id: seed_math_tiger_math_exploit
126
+ config:
127
+ type: load_preexisting
128
+ framework_name: seed_math_tiger_math_exploit
129
+ - id: drop_columns_tiger_math_exploit
130
+ config:
131
+ type: function
132
+ function: data_strategies.commons.keep_only_columns
133
+ function_config:
134
+ columns_to_keep:
135
+ - instruction_seed
136
+ input_ids:
137
+ - seed_math_tiger_math_exploit
138
+ - id: add_source_tiger_math
139
+ config:
140
+ type: function
141
+ function: data_strategies.commons.add_constant_columns
142
+ function_config:
143
+ source: "tiger_math"
144
+ input_ids:
145
+ - drop_columns_tiger_math_exploit
146
+ - id: classify_tiger_math
147
+ config:
148
+ type: fasttext
149
+ num_cpus: 32
150
+ hf_repo_id: mlfoundations-dev/instruction_filtering_fast_text_classifier_seed_math_best_s1
151
+ input_column: instruction_seed
152
+ target_label: "__label__QA_doc"
153
+ top_percentage_ranking: 10
154
+ input_ids:
155
+ - add_source_tiger_math
156
+ - id: sample_dataset_tiger_math
157
+ config:
158
+ type: function
159
+ function: data_strategies.commons.uniform_sample_fixed
160
+ function_config:
161
+ num_samples: 15_000
162
+ input_ids:
163
+ - classify_tiger_math
164
+ - id: fuzzy_dedup
165
+ config:
166
+ type: cpu_function
167
+ num_cpus: 80
168
+ function: data_strategies.DeepSeekR1.utils.fuzzy_dedup
169
+ function_config:
170
+ column: instruction_seed
171
+ similarity_threshold: 90.0
172
+ input_ids:
173
+ - sample_dataset_tiger_math
174
+ - sample_dataset_automath
175
+ - sample_dataset_nvidia_math
176
+ - sample_dataset_tiger_lab_math
177
+ - id: decontaminate
178
+ config:
179
+ type: cpu_function
180
+ num_cpus: 80
181
+ function: data_strategies.DeepSeekR1.utils.decontaminate_table
182
+ function_config:
183
+ column: instruction_seed
184
+ eval_datasets:
185
+ - HuggingFaceH4/MATH-500
186
+ - Maxwell-Jia/AIME_2024
187
+ - AI-MO/aimo-validation-amc
188
+ - livecodebench/code_generation_lite
189
+ - mlfoundations-dev/hero_run_2_fix_conversations
190
+ eval_columns:
191
+ - problem
192
+ - Problem
193
+ - problem
194
+ - question_content
195
+ - question
196
+ eval_splits:
197
+ - test
198
+ - train
199
+ - train
200
+ - test
201
+ - train
202
+ similarity_threshold: 95.0
203
+ input_ids:
204
+ - fuzzy_dedup
205
+ - id: sample_dataset_all
206
+ config:
207
+ type: function
208
+ function: data_strategies.commons.uniform_sample_fixed
209
+ function_config:
210
+ num_samples: 150_000
211
+ input_ids:
212
+ - decontaminate
213
+ - id: add_row_ids
214
+ config:
215
+ type: function
216
+ function: data_strategies.commons.add_row_ids
217
+ function_config:
218
+ column_name: id
219
+ input_ids:
220
+ - sample_dataset_all
221
+ - id: duplicate
222
+ config:
223
+ type: function
224
+ function: data_strategies.TestTime.generator.duplicate_rows
225
+ function_config:
226
+ n_copies: 3
227
+ input_ids:
228
+ - add_row_ids