-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathdata.py
More file actions
251 lines (222 loc) · 9.46 KB
/
data.py
File metadata and controls
251 lines (222 loc) · 9.46 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
import os
from typing import List
import pandas as pd
import numpy as np
from datasets import load_dataset
def create_or_load_large_language_monkeys_mini_f2f_individual_outcomes_df(
raw_data_dir=f"{os.getcwd()}/data/raw_data",
processed_data_dir=f"{os.getcwd()}/data/processed_data",
refresh: bool = False,
) -> pd.DataFrame:
large_language_monkeys_mini_f2f_individual_outcomes_df_path = os.path.join(
processed_data_dir,
"large_language_monkeys_mini_f2f_individual_outcomes.parquet",
)
if refresh or not os.path.exists(
large_language_monkeys_mini_f2f_individual_outcomes_df_path
):
print(
f"Creating {large_language_monkeys_mini_f2f_individual_outcomes_df_path} anew..."
)
os.makedirs(processed_data_dir, exist_ok=True)
large_language_monkeys_original_dfs_list = []
subsets = [
"MiniF2F-MATH_Llama-3-8B-Instruct",
"MiniF2F-MATH_Llama-3-70B-Instruct",
]
for subset in subsets:
benchmark, model = subset.split("_")
ds = load_dataset("ScalingIntelligence/monkey_business", subset)["test"]
correct: List[List[bool]] = ds["is_corrects"]
# Shape: (128, 10000)
wide_df = pd.DataFrame(
correct,
columns=1 + np.arange(10000),
dtype=np.float16,
)
# Convert to floats.
wide_df = wide_df.astype(np.float16)
wide_df["Problem Idx"] = ds["orig_dset_idx"]
df = wide_df.melt(
id_vars=["Problem Idx"],
var_name="Attempt Idx",
value_name="Score",
)
df["Benchmark"] = benchmark
# Convert, e.g., "Pythia-1.4B" to "Pythia 1.4B".
df["Model"] = model.replace("-", " ")
large_language_monkeys_original_dfs_list.append(df)
large_language_monkeys_original_individual_outcomes_df = pd.concat(
large_language_monkeys_original_dfs_list,
)
large_language_monkeys_original_individual_outcomes_df[
"Attempt Idx"
] = pd.to_numeric(
large_language_monkeys_original_individual_outcomes_df["Attempt Idx"]
)
large_language_monkeys_original_individual_outcomes_df.to_parquet(
large_language_monkeys_mini_f2f_individual_outcomes_df_path,
index=False,
)
print(
f"Wrote {large_language_monkeys_mini_f2f_individual_outcomes_df_path} to disk."
)
del large_language_monkeys_original_individual_outcomes_df
large_language_monkeys_original_individual_outcomes_df = pd.read_parquet(
large_language_monkeys_mini_f2f_individual_outcomes_df_path
)
print(
f"Loaded {large_language_monkeys_mini_f2f_individual_outcomes_df_path} with shape: ",
large_language_monkeys_original_individual_outcomes_df.shape,
)
return large_language_monkeys_original_individual_outcomes_df
def create_or_load_large_language_monkeys_code_contests_individual_outcomes_df(
raw_data_dir=f"{os.getcwd()}/data/raw_data",
processed_data_dir=f"{os.getcwd()}/data/processed_data",
refresh: bool = False,
) -> pd.DataFrame:
large_language_monkeys_code_contests_individual_outcomes_df_path = os.path.join(
processed_data_dir,
"large_language_monkeys_code_contests_individual_outcomes.parquet",
)
if refresh or not os.path.exists(
large_language_monkeys_code_contests_individual_outcomes_df_path
):
print(
f"Creating {large_language_monkeys_code_contests_individual_outcomes_df_path} anew..."
)
os.makedirs(processed_data_dir, exist_ok=True)
large_language_monkeys_original_dfs_list = []
subsets = [
"CodeContests_Llama-3-8B",
"CodeContests_Llama-3-8B-Instruct",
"CodeContests_Llama-3-70B-Instruct",
"CodeContests_Gemma-2B",
"CodeContests_Gemma-7B",
]
for subset in subsets:
benchmark, model = subset.split("_")
ds = load_dataset("ScalingIntelligence/monkey_business", subset)["test"]
correct: List[List[bool]] = ds["is_corrects"]
# Shape: (128, 10000)
wide_df = pd.DataFrame(
correct,
columns=1 + np.arange(10000),
dtype=np.float16,
)
# Convert to floats.
wide_df = wide_df.astype(np.float16)
wide_df["Problem Idx"] = ds["orig_dset_idx"]
df = wide_df.melt(
id_vars=["Problem Idx"],
var_name="Attempt Idx",
value_name="Score",
)
df["Benchmark"] = benchmark
# Convert, e.g., "Pythia-1.4B" to "Pythia 1.4B".
df["Model"] = model.replace("-", " ")
large_language_monkeys_original_dfs_list.append(df)
large_language_monkeys_original_individual_outcomes_df = pd.concat(
large_language_monkeys_original_dfs_list,
)
large_language_monkeys_original_individual_outcomes_df[
"Attempt Idx"
] = pd.to_numeric(
large_language_monkeys_original_individual_outcomes_df["Attempt Idx"]
)
large_language_monkeys_original_individual_outcomes_df.to_parquet(
large_language_monkeys_code_contests_individual_outcomes_df_path,
index=False,
)
print(
f"Wrote {large_language_monkeys_code_contests_individual_outcomes_df_path} to disk."
)
del large_language_monkeys_original_individual_outcomes_df
large_language_monkeys_original_individual_outcomes_df = pd.read_parquet(
large_language_monkeys_code_contests_individual_outcomes_df_path
)
print(
f"Loaded {large_language_monkeys_code_contests_individual_outcomes_df_path} with shape: ",
large_language_monkeys_original_individual_outcomes_df.shape,
)
return large_language_monkeys_original_individual_outcomes_df
def create_or_load_large_language_monkeys_pythia_math_individual_outcomes_df(
raw_data_dir=f"{os.getcwd()}/data/raw_data",
processed_data_dir=f"{os.getcwd()}/data/processed_data",
refresh: bool = False,
) -> pd.DataFrame:
large_language_monkeys_pythia_math_individual_outcomes_df_path = os.path.join(
processed_data_dir,
"large_language_monkeys_pythia_math_individual_outcomes.parquet",
)
if refresh or not os.path.exists(
large_language_monkeys_pythia_math_individual_outcomes_df_path
):
print(
f"Creating {large_language_monkeys_pythia_math_individual_outcomes_df_path} anew..."
)
os.makedirs(processed_data_dir, exist_ok=True)
large_language_monkeys_pythia_math_dfs_list = []
subsets = [
"MATH_Pythia-70M",
"MATH_Pythia-160M",
"MATH_Pythia-410M",
"MATH_Pythia-1B",
"MATH_Pythia-2.8B",
"MATH_Pythia-6.9B",
"MATH_Pythia-12B",
]
for subset in subsets:
benchmark, model = subset.split("_")
ds = load_dataset("ScalingIntelligence/monkey_business", subset)["test"]
correct: List[List[bool]] = ds["is_corrects"]
# Shape: (128, 10000)
wide_df = pd.DataFrame(
correct,
columns=1 + np.arange(10000),
dtype=np.float16,
)
# Convert to floats.
wide_df = wide_df.astype(np.float16)
wide_df["Problem Idx"] = ds["orig_dset_idx"]
df = wide_df.melt(
id_vars=["Problem Idx"],
var_name="Attempt Idx",
value_name="Score",
)
df["Benchmark"] = benchmark
# Convert, e.g., "Pythia-1.4B" to "Pythia 1.4B".
df["Model"] = model.replace("-", " ")
large_language_monkeys_pythia_math_dfs_list.append(df)
large_language_monkeys_pythia_math_individual_outcomes_df = pd.concat(
large_language_monkeys_pythia_math_dfs_list,
)
large_language_monkeys_pythia_math_individual_outcomes_df[
"Attempt Idx"
] = pd.to_numeric(
large_language_monkeys_pythia_math_individual_outcomes_df["Attempt Idx"]
)
large_language_monkeys_pythia_math_individual_outcomes_df.to_parquet(
large_language_monkeys_pythia_math_individual_outcomes_df_path,
index=False,
)
print(
f"Wrote {large_language_monkeys_pythia_math_individual_outcomes_df_path} to disk."
)
del large_language_monkeys_pythia_math_individual_outcomes_df
large_language_monkeys_pythia_math_individual_outcomes_df = pd.read_parquet(
large_language_monkeys_pythia_math_individual_outcomes_df_path
)
print(
f"Loaded {large_language_monkeys_pythia_math_individual_outcomes_df_path} with shape: ",
large_language_monkeys_pythia_math_individual_outcomes_df.shape,
)
return large_language_monkeys_pythia_math_individual_outcomes_df
if __name__ == "__main__":
mini_f2f = create_or_load_large_language_monkeys_mini_f2f_individual_outcomes_df()
contests = create_or_load_large_language_monkeys_code_contests_individual_outcomes_df()
pythia_math = create_or_load_large_language_monkeys_pythia_math_individual_outcomes_df()
# breakpoint()
# number of unique "Problem Idx"
# print(pythia_math["Model"].nunique())
# print(mini_f2f.head())