Hugging Face's logo

Datasets:
mlfoundations-dev
/
embedding_apply_nvidia_math

Modalities:
Tabular
Text
Formats:
parquet
Size:
1K - 10K
Libraries:
Datasets
pandas
Dataset card Data Studio Files
xet
 Community
EtashGuha commited on
Commit
607afaf
·
verified ·
1 Parent(s): 2ebb237

Upload configuration file

Browse files
Files changed (1) hide show
  1. config/embedding_apply_nvidia_math.yaml +172 -0
config/embedding_apply_nvidia_math.yaml ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ operators:
2
+ - id: hf_source_numina
3
+ config:
4
+ type: hf_source
5
+ dataset: AI-MO/NuminaMath-CoT
6
+ split: train
7
+ - id: select_amc_aime
8
+ config:
9
+ type: function
10
+ function: data_strategies.commons.select_rows
11
+ function_config:
12
+ conditions:
13
+ source: amc_aime
14
+ input_ids:
15
+ - hf_source_numina
16
+ - id: select_aops
17
+ config:
18
+ type: function
19
+ function: data_strategies.commons.select_rows
20
+ function_config:
21
+ conditions:
22
+ source: aops_forum
23
+ input_ids:
24
+ - hf_source_numina
25
+ - id: rename_task_column_numina
26
+ config:
27
+ type: function
28
+ function: data_strategies.commons.force_rename_columns
29
+ function_config:
30
+ column_maps:
31
+ problem: question
32
+ input_ids:
33
+ - select_amc_aime
34
+ - select_aops
35
+
36
+ - id: hf_source_bad_seed_math
37
+ config:
38
+ type: load_preexisting
39
+ framework_name: seed_math_math_instruct
40
+ - id: rename_task_column
41
+ config:
42
+ type: function
43
+ function: data_strategies.commons.force_rename_columns
44
+ function_config:
45
+ column_maps:
46
+ instruction_seed: question
47
+ input_ids:
48
+ - hf_source_bad_seed_math
49
+
50
+ - id: calc_embeddings_positive
51
+ config:
52
+ type: gpu_function
53
+ num_gpus: 1
54
+ num_shards: 1
55
+ sharded: true
56
+ function: data_strategies.commons.calc_sentence_transformers_embedding
57
+ function_config:
58
+ text_column: question
59
+ model_name: intfloat/multilingual-e5-large-instruct
60
+ input_ids:
61
+ - rename_task_column_numina
62
+
63
+ - id: force_merge_shards_positive
64
+ config:
65
+ type: function
66
+ function: data_strategies.commons.uniform_sample_fixed
67
+ function_config:
68
+ num_samples: 1_000_000
69
+ input_ids:
70
+ - calc_embeddings_positive
71
+
72
+ - id: calc_embeddings_negative
73
+ config:
74
+ type: gpu_function
75
+ num_gpus: 1
76
+ num_shards: 1
77
+ sharded: true
78
+ function: data_strategies.commons.calc_sentence_transformers_embedding
79
+ function_config:
80
+ text_column: question
81
+ model_name: intfloat/multilingual-e5-large-instruct
82
+ input_ids:
83
+ - rename_task_column
84
+
85
+ - id: force_merge_shards_negative
86
+ config:
87
+ type: function
88
+ function: data_strategies.commons.uniform_sample_fixed
89
+ function_config:
90
+ num_samples: 1_000_000
91
+ input_ids:
92
+ - calc_embeddings_negative
93
+
94
+ - id: seed_math_load_in
95
+ config:
96
+ type: load_preexisting
97
+ framework_name: seed_math_load_in
98
+ - id: get_nvidia_math
99
+
100
+ config:
101
+ type: function
102
+ function: data_strategies.commons.select_rows
103
+ function_config:
104
+ conditions:
105
+ source: "nvidia_math"
106
+ input_ids:
107
+ - seed_math_load_in
108
+ - id: sample_nvidia_math
109
+ config:
110
+ type: function
111
+ function: data_strategies.commons.uniform_sample_fixed
112
+ function_config:
113
+ num_samples: 75_000
114
+ input_ids:
115
+ - get_nvidia_math
116
+
117
+ - id: calc_embeddings_nvidia_math
118
+ config:
119
+ type: gpu_function
120
+ num_gpus: 1
121
+ num_shards: 8
122
+ sharded: true
123
+ function: data_strategies.commons.calc_sentence_transformers_embedding
124
+ function_config:
125
+ text_column: instruction_seed
126
+ model_name: intfloat/multilingual-e5-large-instruct
127
+ input_ids:
128
+ - sample_nvidia_math
129
+ - id: force_merge_shards_actual
130
+ config:
131
+ type: function
132
+ function: data_strategies.commons.uniform_sample_fixed
133
+ function_config:
134
+ num_samples: 1_000_000
135
+ input_ids:
136
+ - calc_embeddings_nvidia_math
137
+
138
+
139
+ - id: calc_embedding_difference_nvidia_math
140
+ config:
141
+ type: high_memory_function
142
+ function: data_strategies.InstructionFiltering.EmbeddingFilter.utils.calc_embedding_score_mean
143
+ function_config:
144
+ embedding_column: embeddings
145
+ positive_embedding_column: embeddings
146
+ negative_embedding_column: embeddings
147
+ input_dataset_map:
148
+ dataset: force_merge_shards_actual
149
+ negative_dataset: force_merge_shards_negative
150
+ positive_dataset: force_merge_shards_positive
151
+ input_ids:
152
+ - force_merge_shards_actual
153
+ - force_merge_shards_negative
154
+ - force_merge_shards_positive
155
+ - id: select_to_be_used_nvidia_math
156
+ config:
157
+ type: function
158
+ function: data_strategies.InstructionFiltering.AskLLM.utils.filter_rank
159
+ function_config:
160
+ top_percentage: 0.10
161
+ rating_column: difference_score
162
+ input_ids:
163
+ - calc_embedding_difference_nvidia_math
164
+ - id: sample_dataset_nvidia_math
165
+ config:
166
+ type: function
167
+ function: data_strategies.commons.uniform_sample_fixed
168
+ function_config:
169
+ num_samples: 4_000
170
+ input_ids:
171
+ - select_to_be_used_nvidia_math
172
+