2026-07-08 08:57:50 +00:00

122 lines
3.2 KiB
Markdown

# HellaSwag
## Overview
HellaSwag is a benchmark for evaluating commonsense natural language inference, specifically testing a model's ability to complete sentences describing everyday situations. The dataset uses adversarial filtering to create challenging distractors that are grammatically correct but semantically implausible.
## Task Description
- **Task Type**: Multiple-Choice Sentence Completion
- **Input**: Context describing an activity or situation
- **Output**: Most plausible continuation from 4 choices (A, B, C, D)
- **Domain**: Everyday activities and commonsense scenarios
## Key Features
- 70,000+ questions testing grounded commonsense inference
- Contexts derived from ActivityNet and WikiHow
- Adversarially-filtered incorrect endings
- Requires understanding of typical event sequences
- Tests physical and social commonsense reasoning
## Evaluation Notes
- Default configuration uses **0-shot** evaluation
- Evaluates on the validation split
- Endings are preprocessed to clean formatting artifacts
- Context combines `ctx_a` and `ctx_b` fields
- Activity labels available in metadata for analysis
## Properties
| Property | Value |
|----------|-------|
| **Benchmark Name** | `hellaswag` |
| **Dataset ID** | [evalscope/hellaswag](https://modelscope.cn/datasets/evalscope/hellaswag/summary) |
| **Paper** | N/A |
| **Tags** | `Commonsense`, `Knowledge`, `MCQ` |
| **Metrics** | `acc` |
| **Default Shots** | 0-shot |
| **Evaluation Split** | `validation` |
## Data Statistics
| Metric | Value |
|--------|-------|
| Total Samples | 10,042 |
| Prompt Length (Mean) | 767.6 chars |
| Prompt Length (Min/Max) | 329 / 1655 chars |
## Sample Example
**Subset**: `default`
```json
{
"input": [
{
"id": "18df47e2",
"content": "Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of A,B,C,D.\n\nA man is sitting on a roof. He\n\nA) is using wrap to wrap a pair of skis.\nB) is ripping level tiles off.\nC) is holding a rubik's cube.\nD) starts pulling up roofing on a roof."
}
],
"choices": [
"is using wrap to wrap a pair of skis.",
"is ripping level tiles off.",
"is holding a rubik's cube.",
"starts pulling up roofing on a roof."
],
"target": "D",
"id": 0,
"group_id": 0,
"metadata": {
"activity_label": "Roof shingle removal"
}
}
```
## Prompt Template
**Prompt Template:**
```text
Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}.
{question}
{choices}
```
## Usage
### Using CLI
```bash
evalscope eval \
--model YOUR_MODEL \
--api-url OPENAI_API_COMPAT_URL \
--api-key EMPTY_TOKEN \
--datasets hellaswag \
--limit 10 # Remove this line for formal evaluation
```
### Using Python
```python
from evalscope import run_task
from evalscope.config import TaskConfig
task_cfg = TaskConfig(
model='YOUR_MODEL',
api_url='OPENAI_API_COMPAT_URL',
api_key='EMPTY_TOKEN',
datasets=['hellaswag'],
limit=10, # Remove this line for formal evaluation
)
run_task(task_cfg=task_cfg)
```