File size: 2,535 Bytes
82d75a6 01108bc 82d75a6 0b651c7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 |
---
license: apache-2.0
task_categories:
- text-generation
- text2text-generation
language:
- en
tags:
- code
size_categories:
- n<1K
---
# DafnyBench: A Benchmark for Formal Software Verification
Dataset & code for our paper [DafnyBench: A Benchmark for Formal Software Verification]()
<br>
## Overview 📊
DafnyBench is the largest benchmark of its kind for training and evaluating machine learning systems for formal software verification, with over 750 Dafny programs.
<br><br>
## Usage 💻
- <b>Dataset</b>: The dataset for DafnyBench (with ~782 programs) could be found in the `DafnyBench` directory, which contains the `ground_truth` set & the `hints_removed`set (with compiler hints, i.e. annoataions, removed).
- <b>Evaluation</b>: Evaluate LLMs on DafnyBench by asking models to fill in missing hints in a test file from the `hints_removed` set and checking if the reconstructed program could by verified by Dafny. Please refer to the `eval` directory.
<br>
<p align="center">
<img src="assets/task_overview.jpg" width="600px"/>
</p>
<br><br>
## Set Up for Evaluation 🔧
1. Install Dafny on your machine by following [this tutorial](https://dafny.org/dafny/Installation)
2. Clone & `cd` into this repository
3. Set up environment by running the following lines:
```
python -m venv stats
source stats/bin/activate
pip install -r requirements.txt
cd eval
```
4. Set up environment variable for the root directory:
```
export DAFNYBENCH_ROOT=
```
5. Set up environment variable for path to Dafny executable on your machine (for example, `/opt/homebrew/bin/Dafny`):
```
export DAFNY_PATH=
```
6. If you're evaluating an LLM through API access, set up API key. For example:
```
export OPENAI_API_KEY=
```
7. You can choose to evaluate an LLM on a single test program, such as:
```
python fill_hints.py --model "gpt-4o" --test_file "Clover_abs_no_hints.dfy" --feedback_turn 3 --dafny_path "$DAFNY_PATH"
```
or evaluate on the entire dataset:
```
export model_to_eval='gpt-4o'
./run_eval.sh
```
<br>
## Contents 📁
- `DafnyBench`
- A collection of 782 Dafny programs. Each program has a `ground_truth` version that is fully verified with Dafny & a `hints_removed` version that has hints (i.e. annotations) removed
- `eval`
- Contains scripts to evaluate LLMs on DafnyBench
- `results`
- `results_summary` - Dataframes that summarize LLMs' success on every test program
- `reconstructed_files` - LLM outputs with hints filled back in
- `analysis` - Contains a notebook for analyzing the results
<br> |