Our take
Maintaineda few commits in the last 90 days- Activity
- 33
- Community
- 83
- Issues
- 65
- Pull requests
- 60
+12 stars in the last 30 days
EasyEdit is a broad, actively maintained research toolkit for experimenting with model editing, but its extensive method and model matrix calls for careful compatibility and resource checks.
Good fit if
- You need to compare established knowledge-editing techniques using a shared editor and evaluation workflow.
- Your work involves sequential, safety, or multimodal editing and can use the repository's examples and benchmarks.
- You can work with Python and the model and GPU resources required by your chosen configuration.
Look elsewhere if
- You need a lightweight production service rather than a research framework with model-specific configuration.
- Your target method or model is outside the documented support matrix, or requires one of the methods the README says is unsupported or only partially supported.
- You require an out-of-the-box test and CI setup, which the repository metadata does not report.
All health signals
| Last commit | 2026-10-04 (21 days ago) |
|---|---|
| Commits, last 90 days | 8 |
| Releases, last 12 months | 0 |
| Contributors | 65 (top contributor: 15% of commits) |
| Issues closed, last 90 days | 3 (typically closed in 1 day) |
| Pull requests merged, last 90 days | 3 (typically merged in 0 days) |
| Project age | 3 years |
Checked on 2026-10-26 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
2,943Forks
377Change since Nov 16, 2026: +1 stars·0 forks
Overview
EasyEdit is a Python framework for knowledge editing in large language models, integrating models and tooling from Hugging Face Transformers. It helps researchers and developers change a model's response to specific facts or behaviors while assessing whether the change affects related or unrelated inputs. Its supported scope includes text and multimodal language models.
The project combines editing methods, editor interfaces, and evaluation in one toolkit. It covers single edits as well as sequential edits, and includes datasets and examples for factual updates, safety, personality, conceptual knowledge, and multimodal tasks.
Key Features
- Provides a unified
BaseEditorinterface for applying edits to language models. - Includes a range of editing approaches, including ROME, MEMIT, MEND, IKE, WISE, and AlphaEdit.
- Supports evaluating edit reliability, generalization, locality, portability, and efficiency.
- Supports sequential editing, with rollback documented for GRACE.
- Includes workflows for multimodal editing, safety editing, and personality editing.
- Links to the KnowEdit benchmark and related datasets, including CKnowEdit, SafeEdit, ConceptEdit, and MMEdit.
- Offers example notebooks, configuration files, and training workflows for methods that require pretrained components.
Use Cases
- LLM researchers comparing knowledge-editing methods across edit success and side-effect metrics.
- Teams testing whether a model can be updated with a new fact without changing unrelated answers.
- Safety researchers experimenting with targeted behavior changes and adversarial-input evaluation.
- Multimodal researchers investigating edits to image-captioning or visual-question-answering behavior.
What you need
Detected in the repository
- Python (from requirements.txt)
- A Dockerfile, so it can run in a container
License in plain words
MITpermissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the copyright and license notice
- Share your changes: no
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
The README recommends Python 3.9 or newer. A minimal setup is:
git clone https://github.com/zjunlp/EasyEdit.git
cd EasyEdit
conda create -n EasyEdit python=3.10
conda activate EasyEdit
pip install -r requirements.txt
Then follow the README's method-specific configuration and editing examples: https://github.com/zjunlp/EasyEdit
Alternatives
- KBLaM: KBLaM adds knowledge through trained adapters and special tokens, while EasyEdit focuses on editing a model’s stored knowledge and behavior.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| EasyEdit | Jupyter Notebook | MIT | 2.9k | Maintained |
| KBLaM | Jupyter Notebook | MIT | 1.4k | Maintained |
Considerations
- Editing large models can be resource-intensive. The README's examples report GPU memory use in the tens of gigabytes for several 7B-class model and method combinations; requirements vary by configuration.
- Compatibility is method- and model-dependent. The README documents unsupported methods and notes that MALMEN is only partially supported.
- The project is marked maintained, with recent commits and issue and pull-request activity, but the health data reports no releases in the past year. The README also describes a fast-moving method and model support surface, so check current configurations before adoption.
- The project reports no automated test suite or CI configuration in its repository metadata, which may increase validation work when changing dependencies or extending support.
Found this useful?
Share it with someone who would like EasyEdit.