Update dataset card
Browse files
README.md
CHANGED
@@ -1,33 +1,273 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
dtype: string
|
20 |
-
- name: extracted_at
|
21 |
-
dtype: string
|
22 |
-
splits:
|
23 |
-
- name: train
|
24 |
-
num_bytes: 2366151
|
25 |
-
num_examples: 99
|
26 |
-
download_size: 744933
|
27 |
-
dataset_size: 2366151
|
28 |
configs:
|
29 |
- config_name: default
|
30 |
data_files:
|
31 |
- split: train
|
32 |
-
path: data
|
33 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
license: mit
|
3 |
+
task_categories:
|
4 |
+
- text-classification
|
5 |
+
- feature-extraction
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
tags:
|
9 |
+
- software-engineering
|
10 |
+
- testing
|
11 |
+
- performance
|
12 |
+
- llm-serving
|
13 |
+
- vllm
|
14 |
+
- benchmarking
|
15 |
+
- ml-evaluation
|
16 |
+
pretty_name: vLLM PR Test Classification
|
17 |
+
size_categories:
|
18 |
+
- n<1K
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
configs:
|
20 |
- config_name: default
|
21 |
data_files:
|
22 |
- split: train
|
23 |
+
path: data/*
|
24 |
---
|
25 |
+
|
26 |
+
# vLLM PR Test Classification Dataset
|
27 |
+
|
28 |
+
## ๐ฏ Overview
|
29 |
+
|
30 |
+
This dataset contains **98 vLLM project commits** with their corresponding Pull Request (PR) timeline data and comprehensive test type classifications. It provides insights into testing patterns in a major LLM serving infrastructure project.
|
31 |
+
|
32 |
+
## ๐ Dataset Description
|
33 |
+
|
34 |
+
### Purpose
|
35 |
+
This dataset was created by analyzing vLLM project PR timelines to:
|
36 |
+
- Identify different types of testing and benchmarking activities
|
37 |
+
- Understand testing patterns in LLM infrastructure development
|
38 |
+
- Provide labeled data for ML models to classify test types in software PRs
|
39 |
+
- Enable research on performance optimization trends in LLM serving
|
40 |
+
|
41 |
+
### Test Categories
|
42 |
+
|
43 |
+
Each commit is classified across four test categories:
|
44 |
+
|
45 |
+
| Category | Description | Keywords | Prevalence |
|
46 |
+
|----------|-------------|----------|------------|
|
47 |
+
| **LM Evaluation** | Language model evaluation tests | `lm_eval`, `gsm8k`, `mmlu`, `hellaswag`, `truthfulqa` | 25.5% |
|
48 |
+
| **Performance** | Performance benchmarking tests | `TTFT`, `throughput`, `latency`, `ITL`, `TPOT`, `tok/s` | 81.6% |
|
49 |
+
| **Serving** | Serving functionality tests | `vllm serve`, `API server`, `frontend`, `online serving` | 53.1% |
|
50 |
+
| **General Test** | General testing activities | `CI`, `pytest`, `unittest`, `buildkite`, `fastcheck` | 96.9% |
|
51 |
+
|
52 |
+
## ๐ Dataset Statistics
|
53 |
+
|
54 |
+
### Overall Distribution
|
55 |
+
- **Total commits**: 98
|
56 |
+
- **Multi-category commits**: 76 (77.6%)
|
57 |
+
- **Average test types per commit**: 2.57
|
58 |
+
|
59 |
+
### Detailed Keyword Frequency
|
60 |
+
|
61 |
+
#### Top Performance Keywords (80 commits)
|
62 |
+
- `throughput`: 241 mentions
|
63 |
+
- `latency`: 191 mentions
|
64 |
+
- `profiling`: 114 mentions
|
65 |
+
- `TTFT` (Time To First Token): 114 mentions
|
66 |
+
- `ITL` (Inter-token Latency): 114 mentions
|
67 |
+
- `TPOT` (Time Per Output Token): 108 mentions
|
68 |
+
- `optimization`: 87 mentions
|
69 |
+
- `tok/s` (tokens per second): 66 mentions
|
70 |
+
|
71 |
+
#### Top LM Evaluation Keywords (25 commits)
|
72 |
+
- `gsm8k`: 62 mentions
|
73 |
+
- `lm_eval`: 33 mentions
|
74 |
+
- `lm-eval`: 9 mentions
|
75 |
+
- `mmlu`: 3 mentions
|
76 |
+
- `humaneval`: 1 mention
|
77 |
+
|
78 |
+
#### Top Serving Keywords (52 commits)
|
79 |
+
- `frontend`: 181 mentions
|
80 |
+
- `serving`: 74 mentions
|
81 |
+
- `api server`: 42 mentions
|
82 |
+
- `vllm serve`: 23 mentions
|
83 |
+
- `online serving`: 19 mentions
|
84 |
+
|
85 |
+
## ๐๏ธ Data Schema
|
86 |
+
|
87 |
+
```python
|
88 |
+
{
|
89 |
+
'commit_hash': str, # Git commit SHA-1 hash (40 chars)
|
90 |
+
'pr_url': str, # GitHub PR URL (e.g., https://github.com/vllm-project/vllm/pull/12601)
|
91 |
+
'has_lm_eval': bool, # True if commit contains LM evaluation tests
|
92 |
+
'has_performance': bool, # True if commit contains performance benchmarks
|
93 |
+
'has_serving': bool, # True if commit contains serving tests
|
94 |
+
'has_general_test': bool, # True if commit contains general tests
|
95 |
+
'test_details': str, # Pipe-separated test keywords (e.g., "PERF: ttft, throughput | TEST: ci, pytest")
|
96 |
+
'timeline_text': str, # Full PR timeline text with comments, reviews, and commit messages
|
97 |
+
'extracted_at': str # ISO timestamp when data was extracted
|
98 |
+
}
|
99 |
+
```
|
100 |
+
|
101 |
+
## ๐ป Usage Examples
|
102 |
+
|
103 |
+
### Basic Loading
|
104 |
+
```python
|
105 |
+
from datasets import load_dataset
|
106 |
+
|
107 |
+
# Load the dataset
|
108 |
+
dataset = load_dataset("your-username/vllm-pr-test-classification")
|
109 |
+
|
110 |
+
# Explore the data
|
111 |
+
print(f"Total examples: {len(dataset['train'])}")
|
112 |
+
print(f"Features: {dataset['train'].features}")
|
113 |
+
print(f"First example: {dataset['train'][0]}")
|
114 |
+
```
|
115 |
+
|
116 |
+
### Filtering Examples
|
117 |
+
```python
|
118 |
+
# Get commits with performance benchmarks
|
119 |
+
perf_commits = dataset['train'].filter(lambda x: x['has_performance'])
|
120 |
+
print(f"Performance commits: {len(perf_commits)}")
|
121 |
+
|
122 |
+
# Get commits with LM evaluation
|
123 |
+
lm_eval_commits = dataset['train'].filter(lambda x: x['has_lm_eval'])
|
124 |
+
print(f"LM evaluation commits: {len(lm_eval_commits)}")
|
125 |
+
|
126 |
+
# Get commits with multiple test types
|
127 |
+
multi_test = dataset['train'].filter(
|
128 |
+
lambda x: sum([x['has_lm_eval'], x['has_performance'],
|
129 |
+
x['has_serving'], x['has_general_test']]) >= 3
|
130 |
+
)
|
131 |
+
print(f"Commits with 3+ test types: {len(multi_test)}")
|
132 |
+
```
|
133 |
+
|
134 |
+
### Analysis Example
|
135 |
+
```python
|
136 |
+
import pandas as pd
|
137 |
+
|
138 |
+
# Convert to pandas for analysis
|
139 |
+
df = dataset['train'].to_pandas()
|
140 |
+
|
141 |
+
# Analyze test type combinations
|
142 |
+
test_combinations = df[['has_lm_eval', 'has_performance', 'has_serving', 'has_general_test']]
|
143 |
+
combination_counts = test_combinations.value_counts()
|
144 |
+
print("Most common test combinations:")
|
145 |
+
print(combination_counts.head())
|
146 |
+
|
147 |
+
# Extract performance metrics mentioned
|
148 |
+
perf_df = df[df['has_performance']]
|
149 |
+
print(f"\nCommits mentioning specific metrics:")
|
150 |
+
print(f"TTFT mentions: {perf_df['test_details'].str.contains('TTFT').sum()}")
|
151 |
+
print(f"Throughput mentions: {perf_df['test_details'].str.contains('throughput', case=False).sum()}")
|
152 |
+
```
|
153 |
+
|
154 |
+
### Text Classification Training
|
155 |
+
```python
|
156 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
157 |
+
from transformers import TrainingArguments, Trainer
|
158 |
+
|
159 |
+
# Prepare for multi-label classification
|
160 |
+
def preprocess_function(examples):
|
161 |
+
# Create multi-label targets
|
162 |
+
labels = []
|
163 |
+
for i in range(len(examples['commit_hash'])):
|
164 |
+
label = [
|
165 |
+
int(examples['has_lm_eval'][i]),
|
166 |
+
int(examples['has_performance'][i]),
|
167 |
+
int(examples['has_serving'][i]),
|
168 |
+
int(examples['has_general_test'][i])
|
169 |
+
]
|
170 |
+
labels.append(label)
|
171 |
+
|
172 |
+
# Tokenize timeline text
|
173 |
+
tokenized = tokenizer(
|
174 |
+
examples['timeline_text'],
|
175 |
+
truncation=True,
|
176 |
+
padding='max_length',
|
177 |
+
max_length=512
|
178 |
+
)
|
179 |
+
tokenized['labels'] = labels
|
180 |
+
return tokenized
|
181 |
+
|
182 |
+
# Train a classifier to identify test types from PR text
|
183 |
+
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
|
184 |
+
model = AutoModelForSequenceClassification.from_pretrained(
|
185 |
+
"bert-base-uncased",
|
186 |
+
num_labels=4,
|
187 |
+
problem_type="multi_label_classification"
|
188 |
+
)
|
189 |
+
```
|
190 |
+
|
191 |
+
## ๐ Sample Data
|
192 |
+
|
193 |
+
### Example 1: Performance-focused commit
|
194 |
+
```json
|
195 |
+
{
|
196 |
+
"commit_hash": "fc542144c4477ffec1d3de6fa43e54f8fb5351e8",
|
197 |
+
"pr_url": "https://github.com/vllm-project/vllm/pull/12563",
|
198 |
+
"has_lm_eval": false,
|
199 |
+
"has_performance": true,
|
200 |
+
"has_serving": false,
|
201 |
+
"has_general_test": true,
|
202 |
+
"test_details": "PERF: tok/s, optimization | TEST: CI",
|
203 |
+
"timeline_text": "[Guided decoding performance optimization]..."
|
204 |
+
}
|
205 |
+
```
|
206 |
+
|
207 |
+
### Example 2: Comprehensive testing commit
|
208 |
+
```json
|
209 |
+
{
|
210 |
+
"commit_hash": "aea94362c9bdd08ed2b346701bdc09d278e85f66",
|
211 |
+
"pr_url": "https://github.com/vllm-project/vllm/pull/12287",
|
212 |
+
"has_lm_eval": true,
|
213 |
+
"has_performance": true,
|
214 |
+
"has_serving": true,
|
215 |
+
"has_general_test": true,
|
216 |
+
"test_details": "LM_EVAL: lm_eval, gsm8k | PERF: TTFT, ITL | SERVING: vllm serve | TEST: test, CI",
|
217 |
+
"timeline_text": "[Frontend][V1] Online serving performance improvements..."
|
218 |
+
}
|
219 |
+
```
|
220 |
+
|
221 |
+
## ๐ ๏ธ Potential Use Cases
|
222 |
+
|
223 |
+
1. **Test Type Classification**: Train models to automatically classify test types in software PRs
|
224 |
+
2. **Testing Pattern Analysis**: Study how different test types correlate in infrastructure projects
|
225 |
+
3. **Performance Optimization Research**: Analyze performance testing trends in LLM serving systems
|
226 |
+
4. **CI/CD Insights**: Understand continuous integration patterns in ML infrastructure projects
|
227 |
+
5. **Documentation Generation**: Generate test documentation from PR timelines
|
228 |
+
6. **Code Review Automation**: Build tools to automatically suggest relevant tests based on PR content
|
229 |
+
|
230 |
+
## ๐ Source
|
231 |
+
|
232 |
+
This dataset was extracted from the [vLLM project](https://github.com/vllm-project/vllm) GitHub repository PR timelines. vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.
|
233 |
+
|
234 |
+
## ๐ Updates
|
235 |
+
|
236 |
+
- **v1.0.0** (2025-01): Initial release with 98 commits
|
237 |
+
|
238 |
+
## ๐ License
|
239 |
+
|
240 |
+
This dataset is released under the MIT License, consistent with the vLLM project's licensing.
|
241 |
+
|
242 |
+
## ๐ Citation
|
243 |
+
|
244 |
+
If you use this dataset in your research or applications, please cite:
|
245 |
+
|
246 |
+
```bibtex
|
247 |
+
@dataset{vllm_pr_test_classification_2025,
|
248 |
+
title={vLLM PR Test Classification Dataset},
|
249 |
+
author={vLLM Community Contributors},
|
250 |
+
year={2025},
|
251 |
+
publisher={Hugging Face},
|
252 |
+
url={https://huggingface.co/datasets/your-username/vllm-pr-test-classification},
|
253 |
+
note={A dataset of 98 vLLM commits with test type classifications extracted from GitHub PR timelines}
|
254 |
+
}
|
255 |
+
```
|
256 |
+
|
257 |
+
## ๐ค Contributing
|
258 |
+
|
259 |
+
If you'd like to contribute to this dataset or report issues:
|
260 |
+
1. Open an issue on the Hugging Face dataset repository
|
261 |
+
2. Submit improvements via pull requests
|
262 |
+
3. Share your use cases and findings
|
263 |
+
|
264 |
+
## โ ๏ธ Limitations
|
265 |
+
|
266 |
+
- Dataset size is limited to 98 commits
|
267 |
+
- Timeline text may be truncated for very long PR discussions
|
268 |
+
- Classification is based on keyword matching, which may miss context-dependent references
|
269 |
+
- Dataset represents a snapshot from specific time period of vLLM development
|
270 |
+
|
271 |
+
## ๐ Acknowledgments
|
272 |
+
|
273 |
+
Thanks to the vLLM project maintainers and contributors for their open-source work that made this dataset possible.
|