Init

README.md

@@ -1,2 +1,99 @@
-# BiasEdit
-Debiasing Stereotyped Language Models via Model Editing
+# BiasEdit: Debiasing Stereotyped Language Models via Model Editing
+
+
+<p align="center">
+  <a href="">📃 Paper</a> 
+  <a href="https://github.com/zjunlp/BiasEdit">💻 Code</a> 
+  <a href="">🌏 Web</a> 
+</p>
+
+
+
+<div align=center><img src="fig/BiasEdit_fig1.gif" width="70%"/></div>
+
+
+**BiasEdit** is an efficient *model editing* method to eliminate stereotyped bias from language models with small editor networks, including a *debiasing loss* to guide edits on partial parameters and a *remaining loss* to maintain the language modeling abilities during editing. Experimental results show BiasEdit' excellent performance on debiasing, modeling ability preservation, and robustness of gender reverse and semantic generality.
+
+## 🆕 News
+- [Feb 2024] We released the [paper]() and the refined code.
+- [Dec 2023] Our [idea](https://www.winlp.org/winlp-2023-workshop/accepted-papers/) was accepted by WiNLP 2023 in EMNLP 2023!
+- [Nov 2023] We released the code.
+
+## 📌 Table of Contents
+
+- [🛠️ Setup](#1)
+- [💻 BiasEdit](#2)
+    - [⌚️ Training Editor Networks](#2.1)
+    - [🚀 Debiasing with Editor Networks](#2.2)
+- [👀 Bias Tracing](#3)
+- [📝 Citation](#4)
+- [✨ Acknowledgements](#5)
+
+<h2 id="1">🛠️ Setup</h2>
+
+This codebase uses Python 3.9.18. Other versions may work as well.
+
+Create an environment
+and install the dependencies:
+
+    $ conda create -n biasedit python=3.9
+    $ conda activate biasedit
+    (biasedit) $ pip install -r requirements.txt
+
+
+<h2 id="2">💻 BiasEdit</h2>
+<div align=center><img src="fig/BiasEdit_fig2.png" width="80%"/></div>
+
+With [StereoSet](https://aclanthology.org/2021.acl-long.416/), editor networks are trained to generate parameter shifts for debiasing at first. Then, the trained editor networks are used to conduct edits on language models and produce an unbiased model.
+
+<h3 id="2.1">⌚️ Training Editor Networks</h3>
+
+- Formatted datasets with [train](./data/stereoset/train.json)/[dev](./data/stereoset/dev.json)/test (`gender_test.json`, `race_test.json`, `religion_test.json`) splits are in [data/stereoset](./data/stereoset). 
+- Configurations are in [config](./config). Partial parameters to be edited are presented in [editor](./config/editor). The configurations, like weights to be edited, are in [model](config/model).
+- Experimental scripts are in [scripts](./scripts). All hyper-parameters are in the scripts.
+- For the ablation study on the remaining loss, set `editor.loc_coef=0`.
+- Metrics can be found in the training log.
+
+For example, we use the following command to train the editor networks for Gemma-2B:
+
+
+```bash
+ (biasedit) $ bash scripts/gemma_last2.sh
+```
+
+
+<h3 id="2.2">🚀 Debiasing with Editor Networks</h3>
+
+- Set `eval_only=True`
+- Set `data.valid_path` as the path of the test set
+- Metrics can be found at the end of the debiasing log, like "Test ------- XXX".
+- For testing the robustness of gender reverse, set `data.valid_path` as `data/stereoset/gender_test_reverse.json`.
+- For testing the semantic generality, set `data.valid_path` as `data/stereoset/xxx_test_syn.json`, where `xxx` is chosen from [gender, race, religion].
+
+For example,
+```bash
+ (biasedit) $ bash scripts/gpt2m_last123_gender_reverse.sh
+```
+
+
+<h2 id="3">👀 Bias Tracing</h2>
+
+Enter [bias_tracing](./bias_tracing).
+
+
+<h2 id="4">📝 Citation</h2>
+
+If this code or paper was useful, please consider using the following citation:
+
+    @article{xin24BiasEdit,
+        title={BiasEdit: Debiasing Stereotyped Language Models via Model Editing},
+        author={Xin Xu, Wei Xu, Ningyu Zhang},
+        year={2024},
+        url={https://github.com/zjunlp/BiasEdit}
+    }
+
+<h2 id="5">✨ Acknowledgements</h5>
+
+- Thanks for the original code from [MALMEN](https://github.com/ChenmienTan/malmen).
+- Thanks for StereoSet and all the baselines from [bias-bench](https://github.com/McGill-NLP/bias-bench).
+- For more model editing methods, please try [EasyEdit](https://github.com/zjunlp/EasyEdit).

bias_tracing/README.md

@@ -0,0 +1,28 @@
+# Bias Tracing
+
+Trace bias effect in states of language model.
+
+## Tracing
+Run the scripts `bash scripts/gpt2m.sh`.
+
+Results are saved in `./results`.
+
+## Histograms
+```shell
+>>> python fig.py -h
+    usage: fig.py [-h] [--root ROOT] [--num_layer NUM_LAYER] [--model_name MODEL_NAME] [--bias {gender,race}] [--num_sample NUM_SAMPLE]
+
+    optional arguments:
+    -h, --help            show this help message and exit
+    --root ROOT           the path of results
+    --num_layer NUM_LAYER
+                            The num of model layers.
+    --model_name MODEL_NAME
+                            The model name.
+    --bias {gender,race}  The bias type.
+    --num_sample NUM_SAMPLE
+                            The num of samples
+```
+
+
+Thanks for the original code from [*ROME*](https://github.com/kmeng01/rome).

bias_tracing/dsets/__init__.py

@@ -0,0 +1 @@
+from .stereoset import StereoSetDataset