DeepSeek-Coder/Evaluation/MBPP/utils/dataset.py
2023-11-02 22:07:09 +08:00

49 lines
1.6 KiB
Python

import os
import numpy as np
import json
class MBPPDataset:
def __init__(self, root, samplenum=1):
"""
root: 数据文件的根目录
"""
self.root = root
self.data = open(os.path.join(root, "mbpp.jsonl")).readlines()
self.clean_data = self.get_qa_only_data(self.data)
self.prompt = []
for i in range(1, 4):
prompt = self.clean_data[i]["prompt"]
tests = "\n".join(self.clean_data[i]["test"])
code = self.clean_data[i]["code"].replace("\r", "").replace("\t", " ")
prompt1 = f"You are an expert Python programmer, and here is your task: {prompt} Your code should pass these tests:\n\n{tests}\n[BEGIN]\n{code}\n[DONE]\n"
if len(self.prompt) == 0:
self.prompt.append(prompt1)
else:
self.prompt.append(self.prompt[-1] + prompt1)
self.testdata = []
for i in range(10, 510):
for j in range(samplenum):
self.testdata.append(self.clean_data[i])
np.random.seed(1234)
print(f"Read MBPP from {root}, number of samples {len(self.testdata)}")
def get_qa_only_data(self, data_json):
ans = []
for line in data_json:
line = json.loads(line)
prompt = line["text"]
suffix = line["test_list"]
code = line["code"]
ans.append({"prompt":prompt, "test":suffix, "code":code, "task_id":line["task_id"]})
return ans
def __len__(self):
return len(self.testdata)
def __getitem__(self, index):
sample = self.testdata[index]
return sample