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
|
import os
from OpenCodeEval.benchmark.base import Benchmark, PYTHON_STOP, PYTHON_IMPORTS
from OpenCodeEval.utils import refine_text, stream_jsonl
from OpenCodeEval.eval.func_eval import check_correctness
from OpenCodeEval.eval.sanitize import sanitize
from typing import List, Literal
class mbpp(Benchmark):
name: str = "mbpp"
imports_code = PYTHON_IMPORTS
chat_stop = PYTHON_STOP
base_stop = ['\n"""', "\nassert"]
def __init__(
self,
split: Literal["full", "sanitized"] = "full",
time_out: float = 3.0,
prompt_type: str = "Instruction"
):
super().__init__()
self.split = split
self.time_out = time_out
self.prompt_type = prompt_type
self.path = os.path.join(self.path, f"{self.name}/{self.split}.jsonl")
self.tasks = self.get_task()
self.few_shots_prompt = self.get_few_shots_prompts() if split == "full" else ""
def get_task(self):
"""
Get the task data from the jsonl file into a dictionary.
"""
tasks = {}
for task_data in stream_jsonl(filename = self.path):
task_id = int(task_data["task_id"])
task_data['text'] = refine_text(task_data['text'])
tasks[task_id] = task_data
return tasks
def fewshot_examples(self):
few_shots_start = 1
few_shots_end = 4
few_shots = []
for task_id, task_data in self.tasks.items():
if task_id >= few_shots_start and task_id < few_shots_end:
few_shots.append(task_data)
return few_shots
def format_prompt(self,
promblem: str,
tests: List[str],
code: str = None
) -> str:
promblem = f"You are an expert Python programmer, and here is your task:\n{promblem}"
test = "\n".join(tests)
test = f"Your code should pass these tests:\n{test}\n"
prompt = promblem + test
if code:
code = refine_text(code)
code = f"\n```python\n{code}\n```\n"
prompt = prompt + code
else:
prompt = prompt + "\n```python\n"
return prompt
def get_few_shots_prompts(self):
few_shots_prompts = []
for few_shot in self.fewshot_examples():
few_shots_prompts.append(self.format_prompt(few_shot["text"], few_shot["test_list"], few_shot["code"]))
return '\n'.join(few_shots_prompts)
def get_prompt(self):
"""
Builds the prompt for the LM to generate from.
"""
assert self.prompt_type == "Instruction", "Prompt type must be Instruction for mbpp"
if self.split == "full":
test_start = 10
test_end = 510
elif self.split == "sanitized":
test_start = 0
test_end = 974
prompts = []
for task_id, task_data in self.tasks.items():
if task_id >= test_start and task_id < test_end:
prompt = self.few_shots_prompt + '\n' + self.format_prompt(task_data["text"], task_data["test_list"])
prompts.append({
'task_id': task_id,
'prompt': prompt
})
return prompts
def postprocess_generation(self, generation):
"""
Postprocess the generations.
"""
if generation['completion'].startswith(self.few_shots_prompt):
generation['completion'] = generation['completion'][len(self.few_shots_prompt):]
# if "```python" not in generation['completion']:
# generation['completion'] = ""
return dict(
task_id = generation['task_id'],
completion_id = generation['completion_id'],
solution = sanitize(generation['completion'])
)
def process_results(self, solution):
"""
Takes the list of LM generations and evaluates them against the test cases
"""
task_data = self.tasks[solution['task_id']]
code = (
"\n".join(self.imports_code) + "\n"
+ task_data['test_setup_code'] + "\n"
+ solution['solution'] + "\n"
+ "\n".join(task_data['test_list'])
)
result = check_correctness(solution['task_id'],
solution['completion_id'],
code,
self.time_out)
return result
|
