Improve UltraEditBench dataset card: Add metadata, GitHub link, and sample usage
#2
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,3 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# UltraEditBench
|
| 2 |
|
| 3 |
UltraEditBench is the largest publicly available dataset to date for the task of model editing.
|
|
@@ -6,6 +16,8 @@ This dataset was introduced in the paper:
|
|
| 6 |
|
| 7 |
> [ULTRAEDIT: Training-, Subject-, and Memory-Free Lifelong Editing in Large Language Models](https://arxiv.org/abs/2505.14679)
|
| 8 |
|
|
|
|
|
|
|
| 9 |
---
|
| 10 |
|
| 11 |
## 📦 Dataset Overview
|
|
@@ -44,6 +56,39 @@ Each sample in UltraEditBench includes three full instances (question–answer p
|
|
| 44 |
|
| 45 |
---
|
| 46 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
## 💡 Citation
|
| 48 |
|
| 49 |
If you use this dataset, please cite:
|
|
@@ -60,4 +105,5 @@ If you use this dataset, please cite:
|
|
| 60 |
## 📨 Contact
|
| 61 |
|
| 62 |
- **Email**: [peettherapynoys@gmail.com](mailto:peettherapynoys@gmail.com)
|
| 63 |
-
- **GitHub Issues**: [github.com/XiaojieGu/UltraEdit](https://github.com/XiaojieGu/UltraEdit/issues)
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
task_categories:
|
| 3 |
+
- question-answering
|
| 4 |
+
- text-generation
|
| 5 |
+
language: en
|
| 6 |
+
tags:
|
| 7 |
+
- model-editing
|
| 8 |
+
- lifelong-learning
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
# UltraEditBench
|
| 12 |
|
| 13 |
UltraEditBench is the largest publicly available dataset to date for the task of model editing.
|
|
|
|
| 16 |
|
| 17 |
> [ULTRAEDIT: Training-, Subject-, and Memory-Free Lifelong Editing in Large Language Models](https://arxiv.org/abs/2505.14679)
|
| 18 |
|
| 19 |
+
Code: https://github.com/XiaojieGu/UltraEdit
|
| 20 |
+
|
| 21 |
---
|
| 22 |
|
| 23 |
## 📦 Dataset Overview
|
|
|
|
| 56 |
|
| 57 |
---
|
| 58 |
|
| 59 |
+
## 🚀 Sample Usage
|
| 60 |
+
|
| 61 |
+
### Setup
|
| 62 |
+
Create the environment and install dependencies:
|
| 63 |
+
|
| 64 |
+
```bash
|
| 65 |
+
conda create -n ultraedit python=3.10
|
| 66 |
+
conda activate ultraedit
|
| 67 |
+
pip install torch==2.3.0+cu121 --index-url https://download.pytorch.org/whl/cu121
|
| 68 |
+
pip install -r requirements.txt
|
| 69 |
+
```
|
| 70 |
+
💡 If you want to try editing a Mistral-7B model, even a **24GB consumer GPU** is enough — model editing for everyone!
|
| 71 |
+
|
| 72 |
+
### Run Experiments
|
| 73 |
+
Run the main experiment with:
|
| 74 |
+
|
| 75 |
+
```bash
|
| 76 |
+
sh run.sh
|
| 77 |
+
```
|
| 78 |
+
|
| 79 |
+
The `run.sh` script includes a sample command like:
|
| 80 |
+
|
| 81 |
+
```
|
| 82 |
+
python main.py dataset=zsre model=mistral-7b editor=ultraedit num_seq=200 \ # Number of turns
|
| 83 |
+
editor.cache_dir=cache \
|
| 84 |
+
dataset.batch_size=10 \
|
| 85 |
+
dataset.n_edits=100 \ # Number of edits per turn
|
| 86 |
+
model.edit_modules="[model.layers.29.mlp.down_proj, model.layers.30.mlp.down_proj]"
|
| 87 |
+
```
|
| 88 |
+
💡 Just try editing **20K samples** on Mistral-7B in **under 5 minutes** — ultra-efficient!
|
| 89 |
+
|
| 90 |
+
---
|
| 91 |
+
|
| 92 |
## 💡 Citation
|
| 93 |
|
| 94 |
If you use this dataset, please cite:
|
|
|
|
| 105 |
## 📨 Contact
|
| 106 |
|
| 107 |
- **Email**: [peettherapynoys@gmail.com](mailto:peettherapynoys@gmail.com)
|
| 108 |
+
- **GitHub Issues**: [github.com/XiaojieGu/UltraEdit](https://github.com/XiaojieGu/UltraEdit/issues)
|
| 109 |
+
```
|