Datasets:
Tasks:
Text2Text Generation
Modalities:
Text
Languages:
code
Size:
1K - 10K
ArXiv:
Tags:
code-generation
License:
add daatset script and readme
Browse files- README.md +60 -1
- humaneval_infilling.py +80 -0
README.md
CHANGED
@@ -1,3 +1,62 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
annotations_creators:
|
3 |
+
- expert-generated
|
4 |
+
language_creators:
|
5 |
+
- expert-generated
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
license:
|
9 |
+
- mit
|
10 |
+
multilinguality:
|
11 |
+
- monolingual
|
12 |
+
pretty_name: OpenAI HumanEval-Infilling
|
13 |
+
source_datasets:
|
14 |
+
- original
|
15 |
+
task_categories:
|
16 |
+
- text2text-generation
|
17 |
+
task_ids:
|
18 |
+
- text2text-generation-other-code-generation
|
19 |
---
|
20 |
+
|
21 |
+
# HumanEval-Infilling
|
22 |
+
|
23 |
+
|
24 |
+
## Dataset Description
|
25 |
+
|
26 |
+
- **Homepage:** https://github.com/openai/human-eval-infilling
|
27 |
+
- **Repository:** https://github.com/openai/human-eval-infilling
|
28 |
+
- **Paper:** https://arxiv.org/pdf/2207.14255
|
29 |
+
|
30 |
+
## Dataset Summary
|
31 |
+
|
32 |
+
[HumanEval-Infilling](https://github.com/openai/human-eval-infilling) is a benchmark for infilling tasks, derived from [HumanEval]() benchmark for the evaluation of code generation models.
|
33 |
+
|
34 |
+
## Dataset Structure
|
35 |
+
To load the dataset you need to specify a subset among the 5 exiting languages `[python, cpp, go, java, js]`. By default `python` is loaded.
|
36 |
+
|
37 |
+
```python
|
38 |
+
from datasets import load_dataset
|
39 |
+
ds = load_dataset("humaneval_infilling", "HumanEval-RandomSpanInfilling")
|
40 |
+
|
41 |
+
DatasetDict({
|
42 |
+
test: Dataset({
|
43 |
+
features: ['task_id', 'entry_point', 'prompt', 'suffix', 'canonical_solution', 'test'],
|
44 |
+
num_rows: 1640
|
45 |
+
})
|
46 |
+
})
|
47 |
+
```
|
48 |
+
By default `HumanEval-SingleLineInfilling` subset is loaded.
|
49 |
+
## Subsets
|
50 |
+
|
51 |
+
This dataset has 4 subsets: HumanEval-MultiLineInfilling, HumanEval-SingleLineInfilling, HumanEval-RandomSpanInfilling, HumanEval-RandomSpanInfillingLight.
|
52 |
+
The single-line, multi-line, random span infilling and its light version have 1033, 5815, 1640 and 164 tasks, respectively.
|
53 |
+
|
54 |
+
## Citation
|
55 |
+
|
56 |
+
```@article{bavarian2022efficient,
|
57 |
+
title={Efficient Training of Language Models to Fill in the Middle},
|
58 |
+
author={Bavarian, Mohammad and Jun, Heewoo and Tezak, Nikolas and Schulman, John and McLeavey, Christine and Tworek, Jerry and Chen, Mark},
|
59 |
+
journal={arXiv preprint arXiv:2207.14255},
|
60 |
+
year={2022}
|
61 |
+
}
|
62 |
+
```
|
humaneval_infilling.py
ADDED
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import datasets
|
3 |
+
|
4 |
+
logger = datasets.logging.get_logger(__name__)
|
5 |
+
|
6 |
+
_CITATION = """\
|
7 |
+
@article{bavarian2022efficient,
|
8 |
+
title={Efficient Training of Language Models to Fill in the Middle},
|
9 |
+
author={Bavarian, Mohammad and Jun, Heewoo and Tezak, Nikolas and Schulman, John and McLeavey, Christine and Tworek, Jerry and Chen, Mark},
|
10 |
+
journal={arXiv preprint arXiv:2207.14255},
|
11 |
+
year={2022}
|
12 |
+
}
|
13 |
+
"""
|
14 |
+
|
15 |
+
_DESCRIPTION = """\
|
16 |
+
An evaluation benchamrk for infilling tasks on HumanEval dataset for code generation.
|
17 |
+
"""
|
18 |
+
|
19 |
+
_SUBSETS = [ "MultiLineInfilling", "SingleLineInfilling", "RandomSpanInfilling", "RandomSpanInfillingLight" ]
|
20 |
+
|
21 |
+
|
22 |
+
class HumanevalConfig(datasets.BuilderConfig):
|
23 |
+
"""BuilderConfig for HumanevalConfig."""
|
24 |
+
|
25 |
+
def __init__(
|
26 |
+
self,
|
27 |
+
subset,
|
28 |
+
**kwargs,
|
29 |
+
):
|
30 |
+
self.subset = subset
|
31 |
+
name = f"HumanEval-{subset}"
|
32 |
+
kwargs["name"] = name
|
33 |
+
super(HumanevalConfig, self).__init__(**kwargs)
|
34 |
+
|
35 |
+
|
36 |
+
class MultiPLE(datasets.GeneratorBasedBuilder):
|
37 |
+
BUILDER_CONFIG_CLASS = HumanevalConfig
|
38 |
+
|
39 |
+
BUILDER_CONFIGS = [
|
40 |
+
HumanevalConfig(
|
41 |
+
subset=subset,
|
42 |
+
version=datasets.Version("1.0.0"))
|
43 |
+
for subset in _SUBSETS
|
44 |
+
]
|
45 |
+
|
46 |
+
DEFAULT_CONFIG_NAME = "HumanEval-SingleLineInfilling"
|
47 |
+
|
48 |
+
def _info(self):
|
49 |
+
return datasets.DatasetInfo(
|
50 |
+
description=_DESCRIPTION,
|
51 |
+
license="MIT",
|
52 |
+
features = datasets.Features({'task_id': datasets.Value(dtype='string'),
|
53 |
+
'entry_point': datasets.Value(dtype='string'),
|
54 |
+
'prompt': datasets.Value(dtype='string'),
|
55 |
+
'suffix': datasets.Value(dtype='string'),
|
56 |
+
'canonical_solution': datasets.Value(dtype='string'),
|
57 |
+
'test': datasets.Value(dtype='string')}),
|
58 |
+
supervised_keys=None,
|
59 |
+
homepage="https://github.com/openai/human-eval-infilling",
|
60 |
+
citation=_CITATION
|
61 |
+
)
|
62 |
+
|
63 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager):
|
64 |
+
files = dl_manager.download(
|
65 |
+
f"data/{self.config.name}.jsonl"
|
66 |
+
)
|
67 |
+
return [
|
68 |
+
datasets.SplitGenerator(
|
69 |
+
name=datasets.Split.TEST,
|
70 |
+
gen_kwargs={
|
71 |
+
"filepath": files,
|
72 |
+
}
|
73 |
+
)
|
74 |
+
]
|
75 |
+
|
76 |
+
def _generate_examples(self, filepath):
|
77 |
+
with open(filepath) as f:
|
78 |
+
for id, line in enumerate(f):
|
79 |
+
row = json.loads(line)
|
80 |
+
yield id, row
|