Hugging Face's logo

Datasets:
JetBrains
/
git_good_bench-train

Modalities:
Tabular
Text
Formats:
csv
Size:
10K - 100K
Tags:
git
code
Libraries:
Datasets
pandas
License:
Dataset card Data Studio Files
xet
Community
Liqs commited on
Commit
69930b6
·
verified ·
1 Parent(s): cb0413c

Create dataset fact sheet

Browse files
Files changed (1) hide show
  1. README.md +153 -0
README.md ADDED
@@ -0,0 +1,153 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - git
5
+ - code
6
+ size_categories:
7
+ - 10K<n<100K
8
+ ---
9
+ # Dataset Summary
10
+ GitGoodBench Lite is a subset of 17492 samples for collecting trajectories of AI agents resolving git tasks (see Supported Scenarios) for model training purposes.
11
+ We support the programming languages Python, Java and Kotlin and the sample types merge conflict resolution and file-commit gram.
12
+
13
+ All data in this dataset are collected from 829 unique, open-source GitHub repositories with permissive licenses
14
+ that have >= 1000 stars, >= 5 branches, >= 10 contributors and are not a fork or archived. We collected the initial list of repositories using [SEART.](https://seart-ghs.si.usi.ch/)
15
+
16
+ [For further details see our paper.]()
17
+
18
+ # Supported Tasks
19
+ GitGoodBench Lite contains two types of samples: 'merge' and 'file_commit_chain'.
20
+ It is important to note that the sample type 'file_commit_chain' can be used for two scenario types: Performing an interactive rebase to clean up the local tree or iteratively
21
+ generating commits based on the staged, uncommitted changes.
22
+
23
+ ## Merge
24
+ Merge scenarios are contain one or more merge conflicts that occurred during a merge. All merge conflicts are guaranteed to be in a Python, Java or Kotlin file. There are only
25
+ merges with exactly two parents in our dataset (no octopus merges).
26
+
27
+ A merge scenario looks as follows:
28
+ ```
29
+ {
30
+ 'merge_commit_hash': '9bcf252fb11ec692dfbc152933dddd427098dcc9',
31
+ 'parents': ['5d5df76aa7df56bdbec07c18e063a1125cfd0465', '3bf663778b2a56c614818069043354d4b6d5f156'],
32
+ 'number_of_files_with_merge_conflict': 1,
33
+ 'total_number_of_merge_conflicts': 2,
34
+ 'files_in_merge_conflict': ['models/index_model.py']
35
+ }
36
+ ```
37
+
38
+ Where `merge_commit_hash` contains the ground truth merge commit and the `parents` are the commits during the merge of which the conflict(s) in `files_in_merge_conflict` occurred.
39
+
40
+ ## File-Commit Chain
41
+ File-commit chain scenarios consist of two commits, the oldest and newest commit. In all commits between the `oldest_commit` and `newest_commit` (inclusive) `file` was modified.
42
+ In total the chain consists of `times_seen_consecutively` commits. The intended use-cases of these scenarios are to evaluate the agent's capacity to create meaningful, cohesive commits or
43
+ improve the local tree via rebasing. Thus samples of this `sample_type` cover two scenario types.
44
+
45
+ File-commit chains are at least 3 commits long, all commits only contain files in the specified `programming_language` and no commit is a merge commit.
46
+
47
+ A `file_commit_chain` scenario looks as follows:
48
+ ```
49
+ {
50
+ 'file': 'torchaudio/transforms/_transforms.py',
51
+ 'branch': 'main',
52
+ 'times_seen_consecutively': 3,
53
+ 'purity': 0.69,
54
+ 'newest_commit': '7ac3e2e237e443baf91dfbf9893fca114c1c6001',
55
+ 'oldest_commit': '3742cebb7dc0f8adf24f4ee1cea368195c448f78'
56
+ }
57
+ ```
58
+ `purity` indicates the relative amount of changes in the chain that occurred solely in `file` and is a heuristic for the difficulty of the scenario. We expect noisier scenarios to be more difficult.
59
+
60
+ # Dataset Structure
61
+ The following table provides per-field details. Columns marked “Yes” under **Is Metadata?** are those that provide contextual or descriptive information but are not essential to the primary scenario logic.
62
+
63
+ | **Field** | **Type** | **Description** | **Is Metadata?** |
64
+ |--------------------------|------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------|------------------|
65
+ | **id** | string | A unique identifier for the dataset entry: <name>-<sample_type>-<running_index> | No |
66
+ | **name** | string | The repository name, in “owner/repository” format. | No |
67
+ | **default_branch** | string | The primary or default branch for the repository. | No |
68
+ | **license** | string | Repository license. | Yes |
69
+ | **stargazers** | integer | The number of stars on GitHub. | Yes |
70
+ | **created_at** | string | The repository creation date. | Yes |
71
+ | **topics** | string | A semicolon-delimited list of topics or tags associated with the repository. | Yes |
72
+ | **programming_language** | string | The programming language of the sample. Possible values: "java," "python," or "kotlin." | No |
73
+ | **scenario** | string | A JSON string describing specific scenario data (e.g., merge-conflict details, parent commits). | No |
74
+ | **sample_type** | string | The type of sample. Possible values: "merge" or "file_commit_chain." | No |
75
+ | **project_size** | string | Estimated size based on lines of code. Possible values: "tiny," "small," "medium," "large," or "huge." | Yes |
76
+ | **project_activity** | string | How recently the project was active. Possible values: "day," "week," or "month." | Yes |
77
+ | **difficulty** | string | The complexity level of the scenario. Possible values: "easy," "medium," or "hard." | Yes |
78
+
79
+ **Note**:
80
+ - Fields marked as **Is Metadata? = Yes** provide contextual information (e.g., project stats, licensing) rather than forming the core logic of a scenario.
81
+ - Fields marked **No** represent the primary data for the scenario. Use them to inform or categorize the scenario type and project details.
82
+
83
+ # Dataset statistics
84
+ We provide some statistics on the distribution of “difficulty” within the overall dataset and across different scenario types and on the distribution of samples across programming languages and
85
+ the skew of our dataset.
86
+
87
+ ## Dataset Skew
88
+
89
+ The below statistics show that our dataset does not exhibit an extreme skew towards some repositories and is relatively well balanced with respect to source repositories.
90
+
91
+ ### Distribution Statistics
92
+ - Total number of repositories analyzed: 829
93
+ - Average (mean) samples per repository: 21.10
94
+ - Standard deviation (std): 48.46
95
+ - Minimum (min): 1
96
+ - 25th percentile (25%): 2
97
+ - Median (50%): 6
98
+ - 75th percentile (75%): 18
99
+ - Maximum (max): 645
100
+
101
+ ### Top-10 Repositories by Sample Count
102
+
103
+ | Repository | Percentage of Total Samples |
104
+ |--------------------------------|----------------------------:|
105
+ | zulip/zulip | 3.69% |
106
+ | trinodb/trino | 2.46% |
107
+ | wandb/wandb | 2.45% |
108
+ | facebook/litho | 2.15% |
109
+ | oss-review-toolkit/ort | 2.03% |
110
+ | apache/tomcat | 1.94% |
111
+ | nvidia/nemo | 1.76% |
112
+ | h2oai/h2ogpt | 1.30% |
113
+ | conan-io/conan | 1.30% |
114
+ | huggingface/transformers | 1.05% |
115
+
116
+
117
+ ### Distribution of Programming Languages
118
+
119
+ We do however note a severe skew towards Python and Java with only 3.8% of samples being Kotlin.
120
+
121
+ | Programming Language | Count | Percentage |
122
+ |----------------------|--------:|-----------:|
123
+ | python | 10956 | 62.63% |
124
+ | java | 5871 | 33.56% |
125
+ | kotlin | 665 | 3.80% |
126
+
127
+
128
+ ## Overall Difficulty Distribution
129
+
130
+ | Difficulty | Proportion |
131
+ |------------|-----------:|
132
+ | easy | 0.409845 |
133
+ | hard | 0.337240 |
134
+ | medium | 0.252916 |
135
+
136
+ ## Difficulty Distribution for "merge" Scenarios
137
+
138
+ | Difficulty | Proportion |
139
+ |------------|-----------:|
140
+ | easy | 0.516466 |
141
+ | hard | 0.299672 |
142
+ | medium | 0.183861 |
143
+
144
+ ## Difficulty Distribution for "file_commit_chain" Scenarios
145
+
146
+ | Difficulty | Proportion |
147
+ |------------|-----------:|
148
+ | hard | 0.358954 |
149
+ | easy | 0.348218 |
150
+ | medium | 0.292828 |
151
+
152
+
153
+ **Languages** We note that the text data in this dataset consists mostly of: commit messages, comments and is primarily in English. We do however not filter for any human languages explcitly.