-
Notifications
You must be signed in to change notification settings - Fork 2
/
llm_eval_harness.py
173 lines (141 loc) · 5.26 KB
/
llm_eval_harness.py
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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Adapted from https://github.com/bjoernpl/lm-evaluation-harness-de/blob/mmlu_de/eval_de.py
"""
import os
import argparse
import json
import logging
from pathlib import Path
import random
from time import time
from lm_eval import tasks, evaluator, utils
import lm_eval.models
import numpy as np
import pandas as pd
logging.getLogger("openai").setLevel(logging.WARNING)
os.environ["PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION"] = "python"
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--model", required=True)
parser.add_argument("--model_args", default="")
parser.add_argument("--tasks", default=None, choices=utils.MultiChoice(tasks.ALL_TASKS))
parser.add_argument("--provide_description", action="store_true")
parser.add_argument("--num_fewshot", type=int, default=0)
parser.add_argument("--batch_size", type=int, default=4)
parser.add_argument("--device", type=str, default=None)
parser.add_argument("--limit", type=float, default=None,
help="Limit the number of examples per task. "
"If <1, limit is a percentage of the total number of examples.")
parser.add_argument("--data_sampling", type=float, default=None)
parser.add_argument("--no_cache", action="store_true")
parser.add_argument("--decontamination_ngrams_path", default=None)
parser.add_argument("--description_dict_path", default=None)
parser.add_argument("--check_integrity", action="store_true")
parser.add_argument("--write_out", action="store_true", default=False)
parser.add_argument("--output_base_path", type=str, default="resources/outputs/lm_eval_harness")
parser.add_argument("--skip_fewshots", type=list, default=[])
parser.add_argument("--debug", action="store_true")
return parser.parse_args()
tasks_per_fewshot = {
# 5: [
# "hendrycksTest*",
# "MMLU-DE*",
# ],
# 10: [
# "hellaswag",
# "hellaswag_de"
# ],
# 25: [
# "arc_challenge",
# "arc_challenge_de"
# ],
0: [
"xwinograd_en", # ["en", "fr", "jp", "pt", "ru", "zh"]
"xwinograd_fr",
"xwinograd_jp",
"xwinograd_pt",
"xwinograd_ru",
"xwinograd_zh",
"pawsx_en", # ["en", "de", "es", "fr", "ja", "ko", "zh",]
"pawsx_de",
"pawsx_es",
"pawsx_fr",
"pawsx_ja",
"pawsx_ko",
"pawsx_zh",
"xnli_en", # ["ar", "bg", "de", "el", "en", "es", "fr", "hi", "ru", "sw", "th", "tr", "ur", "vi", "zh",]
"xnli_de",
"xnli_es",
"xnli_fr",
"xnli_ru",
"xnli_zh",
"xnli_bg",
"xnli_el",
"xnli_hi",
"xnli_vi",
]
}
def main():
args = parse_args()
# create output path
Path(args.output_base_path).mkdir(parents=True, exist_ok=True)
all_results = {
"config": {"model": args.model, "model_args": args.model_args},
"results": {},
"versions": {},
}
model_name = Path(args.model_args.split(',')[0].split('=')[1]).name
for num_fewshots, task_list in tasks_per_fewshot.items():
start = time()
task_names = utils.pattern_match(task_list, tasks.ALL_TASKS)
if args.debug:
random.seed(0)
task_names = random.sample(task_names, 4)
print(
f"Running:\n"
f"{args.model} ({args.model_args})\n"
f"limit: {args.limit}\n"
f"provide_description: {args.provide_description}\n"
f"num_fewshot: {args.num_fewshot}\n"
f"batch_size: {args.batch_size}\n"
f"device: {args.device}\n"
f"no_cache: {args.no_cache}\n"
f"tasks: {task_names}\n"
)
results = evaluator.simple_evaluate(
model=args.model,
model_args=args.model_args,
tasks=task_names,
num_fewshot=num_fewshots,
batch_size=args.batch_size,
device=args.device,
no_cache=args.no_cache,
limit=args.limit,
description_dict=None,
decontamination_ngrams_path=args.decontamination_ngrams_path,
check_integrity=args.check_integrity,
write_out=args.write_out,
output_base_path=args.output_base_path,
bootstrap_iters=100000
)
all_results["results"].update(results["results"])
all_results["versions"].update(results["versions"])
all_results["config"] = results["config"]
time_taken = time() - start
all_results["time_taken"] = time_taken
dumped = json.dumps(all_results, indent=2)
print(dumped)
output_file = Path(args.output_base_path) / f'{model_name}_fs{num_fewshots}.json'
with open(output_file, "w", encoding='utf8') as f:
f.write(dumped)
print(f'Wrote results {num_fewshots}-shot results to {output_file} in {time_taken} seconds.')
dumped = json.dumps(all_results, indent=2)
print(dumped)
output_file = Path(args.output_base_path) / f'{model_name}.json'
with open(output_file, "w", encoding='utf8') as f:
f.write(dumped)
print(f'Wrote results to {output_file}')
if __name__ == "__main__":
main()