![](https://raw.githubusercontent.com/AllYoung/EasyEdit/main/figs/logo.png)
An Easy-to-use Knowledge Editing Framework for Large Language Models.
Installation • QuickStart • Doc • Paper • Demo • Benchmark • Contributors • Slides • Video • Featured By AK
- Table of Contents
- 🔔News
- Editing Demo
- Knowledge Editing
- 🌟Overview
- Requirements
- 📌Use EasyEdit
- Use EasyEdit with KnowEdit
- Editing Performance
- Citation
- 🎉Contributors
- 2024-03-22 We release a new paper:"Detoxifying Large Language Models via Knowledge Editing" with a new dataset SafeEdit and a new detoxification method DINM!
- 2024-03-12 We release a new paper:"Editing Conceptual Knowledge for Large Language Models" with a new dataset ConceptEdit!
- 2024-03-01 The EasyEdit has added the support for a new method called FT-M, which trains the specific MLP layer using the cross-entropy loss on the target answer and masks the origin text. This method achieves better performance than the FT-L implementation in ROME. We thank the author of the issue zjunlp#173 for their advice.
- 2024-02-27 The EasyEdit has added the support for a new method called InstructEdit and technical details are in the paper "InstructEdit: Instruction-based Knowledge Editing for Large Language Models".
- 2024-02-20 The AAAI2024 tutorial "Knowledge Editing for Large Language Models" has been canceled since speakers cannot present in person, we make this ppt[Github] [Google Drive] [Baidu Pan] available to the community.
- 2024-02-09 The EasyEdit has added the support for the Dynamic LoRA model editing method MELO'AAAI24.
- 2024-02-06 We release a new paper: "EasyInstruct: An Easy-to-use Instruction Processing Framework for Large Language Models" with an HF demo EasyInstruct.
- 2024-02-06 We release a preliminary tool EasyDetect for LLM hallucination detection,with a demo.
- 2024-01-24 The EasyEdit has added the support for editing Mistral-7B (manually update transformers==4.34.0), we have also fixed some bugs in evaluating MEND (slightly influence the performance).
- 2024-01-16 The EasyEdit has added the support for the precise model editing method PMET'AAAI24.
- 2024-01-03 We release a new paper:"A Comprehensive Study of Knowledge Editing for Large Language Models" with a new benchmark KnowEdit! We are looking forward to any comments or discussions on this topic :)
Previous News
- 2023-12-06 The EasyEdit has added the support for the lifelong model editing method GRACE'NeurIPS24.
- 2023-11-18 Our tutorial "Knowledge Editing for Large Language Models" has been accepted by COLING 2024.
- 2023-10-25 Our tutorial "Knowledge Editing for Large Language Models" has been accepted by AAAI 2024.
- 2023-10-24 The EasyEdit has added the support for efficient editing of Baichuan2, ChatGLM2, InternLM, Qwen and fixed several bugs for a better user experience.
- 2023-10-14 We release the MultimodalEditor based on the paper "Can We Edit Multimodal Large Language Models?".
- 2023-10-13 We release the paper "Can We Edit Multimodal Large Language Models?" accepted by EMNLP 2023.
- 2023-10-08 Our paper "Editing Large Language Models: Problems, Methods, and Opportunities" has been accepted by EMNLP 2023.
- 2023-10-07 The EasyEdit has added the support for editing models with multiple GPUs, using huggingface
Accelerate
. - 2023-9-21 The EasyEdit has added the support for Parameter-Efficient Fine-Tuning through AdaLoRA to inject knowledge into the LLM.
- 2023-8-31 The EasyEdit has added the support for official fine-tuning API for gpt-3.5-turbo to customize ChatGPT for your editing cases.
- 2023-8-15 We release the paper "EasyEdit: An Easy-to-use Knowledge Editing Framework for Large Language Models."
- 2023-7-12 We release version 0.0.1, supporting several knowledge editing techniques for LLMs. EasyEdit helps to better align LLMs with changing needs and values of users.
- 2023-5-22 We release the paper "Editing Large Language Models: Problems, Methods, and Opportunities" and provide a paper list at PaperList.
- 2023-3-25 The EasyEdit project has been launched and is under development.
This repository is a subproject of KnowLM.
A Comprehensive Study of Knowledge Editing for Large Language Models [paper][benchmark][code]
AAAI 2024 Tutorial [Google Drive] [Baidu Pan]
AACL 2023 Tutorial [Google Drive] [Baidu Pan]
There is a demonstration of editing. The GIF file is created by Terminalizer.
Deployed models may still make unpredictable errors. For example, Large Language Models (LLMs) notoriously hallucinate, perpetuate bias, and factually decay, so we should be able to adjust specific behaviors of pre-trained models.
Knowledge editing aims to adjust an initial base model's
Inject knowledge that LLMs have not seen before. such as:
-
How many times has Messi won the World Cup? 0
$\rightarrow$ 1:-
$x_e$ : How many times has Messi won the World Cup?$\quad$ $y_e$ : 1
-
LLMs often suffer from knowledge cutoff issue, EasyEdit can update outdated knowledge. such as:
-
The president of USA: Donald Trump
$\rightarrow$ Joe Biden:-
$x_e$ : Who is the president of the US?$\quad$ $y_e$ : Joe Biden
-
EasyEdit can erase sensitive information. such as:
-
The phone number of someone is XXXX
$\rightarrow$ __-
$x_e$ : The phone number of someone is$\quad$ $y_e$ : __
-
Without influencing the model behavior on unrelated samples, the ultimate goal is to create an edited model
The knowledge editing process generally impacts the predictions for a broad set of inputs that are closely associated with the edit example, called the editing scope.
A successful edit should adjust the model’s behavior within the editing scope while remaining unrelated inputs(as below formula).
In addition to this, the performance of knowledge editing should be measured from multiple dimensions:
Reliability
: the success rate of editing with a given editing descriptionGeneralization
: the success rate of editing within the editing scopeLocality
: whether the model's output changes after editing for unrelated inputsPortability
: the success rate of editing for factual reasoning(one hop, synonym, one-to-one relation)Efficiency
: time and memory consumption required during the editing process
EasyEdit is a Python package for edit Large Language Models (LLM) like GPT-J
, Llama
, GPT-NEO
, GPT2
, T5
(support models from 1B to 65B), the objective of which is to alter the behavior of LLMs efficiently within a specific domain without negatively impacting performance across other inputs. It is designed to be easy to use and easy to extend.
-
EasyEdit contains a unified framework for Editor, Method and Evaluate, respectively representing the editing scenario, editing technique, and evaluation method.
-
Each Knowledge Editing scenario comprises of three components:
-
Editor
: such as BaseEditor(Factual Knowledge and Generation Editor) for LM, MultiModalEditor(MultiModal Knowledge). -
Method
: the specific knowledge editing technique used(such as ROME, MEND, ..). -
Evaluate
: Metrics for evaluating knowledge editing performance.-
Reliability
,Generalization
,Locality
,Portability
-
-
-
The current supported knowledge editing techniques are as follows:
-
FT: Fine-Tuning with
$L_\infty$ constraint (FT-L) or finetuning using the cross-entropy loss on the target output (FT-M), both can be found in the hparams. - SERAC: Mitchell et al. Memory-based
- IKE: Ce Zheng et al. In-Context Editing
- MEND: Mitchell et al. Hypernetwork
- KN: Damai Dai et al. Locate then Edit
- ROME: Kevin Meng et al. Locate and Edit
- MEMIT: Kevin Meng et al. Locate and Edit
- GRACE: Thomas Hartvigsen et al. Memory-based
- MELO: Lang Yu et al. Memory-based
- PMET: Xiaopeng Li et al. Locate and Edit
- MALMEN: Chenmien Tan et al. Hypernetwork
-
InstructEdit: Bozhong Tian et al. Hypernetwork
Due to the limited compatibility of this toolkit and limited by the transformer version, some knowledge editing methods including T-Patcher, KE, CaliNet are not supported. Similarly, the MALMEN method is only partially supported due to the same reasons and will continue to be improved upon in the first half of the year.
-
FT: Fine-Tuning with
You can choose different editing methods according to your specific needs.
Method | T5 | GPT-2 | GPT-J | GPT-NEO | LlaMA | Baichuan | ChatGLM2 | InternLM | Qwen | Mistral |
---|---|---|---|---|---|---|---|---|---|---|
FT | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
AdaLoRA | ✅ | |||||||||
SERAC | ✅ | ✅ | ✅ | ✅ | ||||||
IKE | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
MEND | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
KN | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | |
ROME | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | |
MEMIT | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | |
GRACE | ✅ | ✅ | ✅ | |||||||
MELO | ✅ | |||||||||
PMET | ✅ | ✅ | ||||||||
InstructEdit | ✅ | ✅ |
❗️❗️ EasyEdit supports editing ChatGPT with FT. An edit for
gpt-3.5-turbo
returns model_name(for example,ft: GPT-3.5-turbo-0613 :personal::7tWZkLzq
) instead model weights.
❗️❗️ If you intend to use Mistral, please update the
transformers
library to version 4.34.0 manually. You can use the following code:pip install transformers==4.34.0
.
❗️❗️ If you intend to use MELO, please get the in ./easyeditor/models/melo/peft_egg and pip install it in your environment.
Benchmark: KnowEdit [Hugging Face][WiseModel][ModelScope]
Task | Knowledge Insertion | Knowledge Modification | Knowledge Erasure | |||
---|---|---|---|---|---|---|
Datasets | Wikirecent | ZsRE | WikiBio | WikiDatacounterfact | Convsent | Sanitation |
Type | Fact | Question Answering | Hallucination | Counterfact | Sentiment | Unwanted Info |
# Train | 570 | 10,000 | 592 | 1,455 | 14,390 | 80 |
# Test | 1,266 | 1230 | 1,392 | 885 | 800 | 80 |
We provide detailed scripts for user to easily use KnowEdit, please refer to examples.
dataset description
- ZsRE: is a context-free question-answering task. Given a question based on the subject and relation, the model is expected to provide the correct object as the answer.
- Wikirecent: This dataset specifically focuses on triplets that have been recently inserted into WikiData after July 2022.
- WikiBio: The original dataset was created by prompting GPT-3 to generate 238 Wikipedia-style biographies using subjects from the WikiBio.
- WikiDatacounterfact: Since tail entities are often not captured by models, and therefore are not suitable for testing modification edits, RippleEdit collects triplets about popular entities, where the subject corresponds to one of the top-viewed pages in Wikipedia.
- Convsent: This is a sentiment editing task that assesses the model's ability to modify a dialog agent's sentiment on a specific topic without affecting its responses to other topics.
- Sanitation: This dataset specifically addresses privacy concerns associated with learned language models.
dataset structure
knowedit
├── WikiBio
│ ├── wikibio-test-all.json
│ └── wikibio-train-all.json
├── ZsRE
│ └── ZsRE-test-all.json
├── wiki_counterfact
│ ├── test_cf.json
│ └── train_cf.json
├── convsent
│ ├── blender_test.json
│ ├── blender_train.json
│ └── blender_val.json
├── convsent
│ ├── trivia_qa_test.json
│ └── trivia_qa_train.json
└── wiki_recent
├── recent_test.json
└── recent_train.json
dataset | Google Drive | BaiduNetDisk | Description |
---|---|---|---|
ZsRE plus | [Google Drive] | [BaiduNetDisk] | Question Answering dataset using question rephrasings |
Counterfact plus | [Google Drive] | [BaiduNetDisk] | Counterfact dataset using Entity replacement |
We provide zsre and counterfact datasets to verify the effectiveness of knowledge editing. You can download them here. [Google Drive], [BaiduNetDisk].
- For locality, in addition to testing unrelated instances, we also provide tests on distracting (reference: Detecting Edit Failures...), other attribution, and other downstream tasks (such as commonsense reasoning).
- For portability, it tests whether the model can apply edited instances for inference. We provide evaluations for one-hop reasoning, subject alias, and inverse relation (eg, a one-to-one relationship between spouses should be bidirectionally edited).
dataset description
editing-data
├── counterfact
│ ├── counterfact-edit.json
│ ├── counterfact-train.json
│ └── counterfact-val.json
├── locality
│ ├── Commonsense Task
│ │ ├── piqa_valid-labels.lst
│ │ └── piqa_valid.jsonl
│ ├── Distracting Neighbor
│ │ └── counterfact_distracting_neighbor.json
│ └── Other Attribution
│ └── counterfact_other_attribution.json
├── portability
│ ├── Inverse Relation
│ │ └── zsre_inverse_relation.json
│ ├── One Hop
│ │ ├── counterfact_portability_gpt4.json
│ │ └── zsre_mend_eval_portability_gpt4.json
│ └── Subject Replace
│ ├── counterfact_subject_replace.json
│ └── zsre_subject_replace.json
└── zsre
├── zsre_mend_eval.json
├── zsre_mend_train_10000.json
└── zsre_mend_train.json
- counterfact: original counterfact dataset using Entity replacement
- zsre: original question answering dataset using question rephrasings
- locality (evaluation for locality, see details in this paper)
- Commonsense Task: evaluation for other downstream tasks such as commonsense task
- Distracting Neighbor: test on distracting neighborhood (reference: Detecting Edit Failures...)
- Other Attribution
- portability
- Inverse Relation: evaluation for one-to-one relationship such as
spouse
- One Hop: evaluation for one-hop reasoning
- Subject Replace: evaluation for synonym replacement
- Inverse Relation: evaluation for one-to-one relationship such as
dataset | Google Drive | HuggingFace Dataset | Description |
---|---|---|---|
ConceptEdit | [Google Drive] | [HuggingFace Dataset] | dataset for editing conceptual knowledge |
- Here, you can follow ConceptEdit.md to run concept editing experiments.
dataset description
data
└──concept_data.json
├──final_gpt2_inter.json
├──final_gpt2_intra.json
├──final_gptj_inter.json
├──final_gptj_intra.json
├──final_llama2chat_inter.json
├──final_llama2chat_intra.json
├──final_mistral_inter.json
└──final_mistral_intra.json
Concept Specific Evaluation Metrics
Instance Change
: capturing the intricacies of these instance-level changesConcept Consistency
: the semantic similarity of generated concept definition
dataset | Google Drive | BaiduNetDisk | Description |
---|---|---|---|
E-IC | [Google Drive] | [BaiduNetDisk] | dataset for editing Image Captioning |
E-VQA | [Google Drive] | [BaiduNetDisk] | dataset for editing Visual Question Answering |
- All images used in E-IC and E-VQA are available for download at Google Drive
- For locality, it is the same as factual editing in order to measure whether unrelated facts retain their outputs.
- For multimodal locality, it assesses the impact of editing on the visual module, which is similar to regular locality.
dataset description
editing-data
├── caption
│ ├── caption_train_edit.json
│ └── caption_eval_edit.json
├── locality
│ ├── NQ dataset
│ │ ├── train.json
│ │ └── validation.json
├── multimodal_locality
│ ├── OK-VQA dataset
│ │ ├── okvqa_loc.json
└── vqa
├── vqa_train.json
└── vqa_eval.json
- Multimodal locality (evaluation for multimodal locality, see dataset's details in this paper)
dataset | HuggingFace Dataset | Description |
---|---|---|
SafeEdit | [HuggingFace Dataset] | dataset for detoxifying LLMs |
- Here, you can follow SafeEdit.md to run detoxification editing experiments.
dataset description
data
└──SafeEdit_train.json
└──SafeEdit_val.json
└──SafeEdit_test.json
Detoxifying Specific Evaluation Metrics
Defense Duccess (DS)
: the detoxification success rate of edited LLM for adversarial input (attack prompt + harmful question), which is used to modify LLM.Defense Generalization (DG)
: the detoxification success rate of edited LLM for out-of-domain malicious inputs.General Performance
: the side effects for unrelated task performance.
Method | Description | GPT-2 | LlaMA |
---|---|---|---|
IKE | In-Context Learning (ICL) Edit | [Colab-gpt2] | [Colab-llama] |
ROME | Locate-Then-Edit Neurons | [Colab-gpt2] | [Colab-llama] |
MEMIT | Locate-Then-Edit Neurons | [Colab-gpt2] | [Colab-llama] |
Note: Please use Python 3.9+ for EasyEdit To get started, simply install conda and run:
git clone https://github.com/zjunlp/EasyEdit.git
conda create -n EasyEdit python=3.9.7
...
pip install -r requirements.txt
We packaged the environment, you can download Docker from this link.
Pull the Docker image from Docker Hub or Aliyun:
docker pull zjunlp/easyedit
docker pull registry.cn-hangzhou.aliyuncs.com/zjunlp/easyedit:v1
If you want to build the Docker image locally, you can clone the project to your local machine and build the Docker image:
git clone https://github.com/zjunlp/EasyEdit.git
cd EasyEdit
docker build -t your-image-name .
Then run the Docker image as a container:
docker run -p 8080:80 your-image-name
Our results are all based on the default configuration
llama-2-7B | chatglm2 | gpt-j-6b | gpt-xl | |
---|---|---|---|---|
FT | 60GB | 58GB | 55GB | 7GB |
SERAC | 42GB | 32GB | 31GB | 10GB |
IKE | 52GB | 38GB | 38GB | 10GB |
MEND | 46GB | 37GB | 37GB | 13GB |
KN | 42GB | 39GB | 40GB | 12GB |
ROME | 31GB | 29GB | 27GB | 10GB |
MEMIT | 33GB | 31GB | 31GB | 11GB |
AdaLoRA | 29GB | 24GB | 25GB | 8GB |
GRACE | 27GB | 23GB | 6GB |
-
Edit large language models(LLMs) around 5 seconds
-
Following example shows you how to perform editing with EasyEdit. More examples and tutorials can be found at examples
BaseEditor
is the class for Language Modality Knowledge Editing. You can choose the appropriate editing method based on your specific needs.
- Due to different transformer versions and different GPU models, the editing results may fluctuate slightly.
With the modularity and flexibility of EasyEdit
, you can easily use it to edit model.
Step1: Define a PLM as the object to be edited.
Choose the PLM to be edited. EasyEdit
supports partial models(T5
, GPTJ
, GPT-NEO
, LlaMA
so far) retrievable on HuggingFace. The corresponding configuration file directory is hparams/YUOR_METHOD/YOUR_MODEL.YAML
, such as hparams/MEND/gpt2-xl.yaml
, set the corresponding model_name
to select the object for knowledge editing.
model_name: gpt2-xl
model_class: GPT2LMHeadModel
tokenizer_class: GPT2Tokenizer
tokenizer_name: gpt2-xl
model_parallel: false # true for multi-GPU editing
Step2: Choose the appropriate Knowledge Editing Method The selection of editing methods is a crucial step, as different methods have their own strengths and weaknesses. Users need to consider the trade-off between editing success rate, generalization, and maintaining unrelated performance. For specific performance details of each method, please refer to the paper: Editing Large Language Models: Problems, Methods, and Opportunities.
## In this case, we use MEND method, so you should import `MENDHyperParams`
from easyeditor import MENDHyperParams
## Loading config from hparams/MEMIT/gpt2-xl.yaml
hparams = MENDHyperParams.from_hparams('./hparams/MEND/gpt2-xl')
Step3: Provide the edit descriptor and edit target
## edit descriptor: prompt that you want to edit
prompts = [
'What university did Watts Humphrey attend?',
'Which family does Ramalinaceae belong to',
'What role does Denny Herzig play in football?'
]
## You can set `ground_truth` to None !!!(or set to original output)
ground_truth = ['Illinois Institute of Technology', 'Lecanorales', 'defender']
## edit target: expected output
target_new = ['University of Michigan', 'Lamiinae', 'winger']
Step4: Combine them into a BaseEditor
EasyEdit
provides a simple and unified way to init Editor, like huggingface: from_hparams.
## Construct Language Model Editor
editor = BaseEditor.from_hparams(hparams)
Step5: Provide the data for evaluation Note that the data for portability and locality are both optional(set to None for basic editing success rate evaluation only). The data format for both is a dict, for each measurement dimension, you need to provide the corresponding prompt and its corresponding ground truth. Here is an example of the data:
locality_inputs = {
'neighborhood':{
'prompt': ['Joseph Fischhof, the', 'Larry Bird is a professional', 'In Forssa, they understand'],
'ground_truth': ['piano', 'basketball', 'Finnish']
},
'distracting': {
'prompt': ['Ray Charles, the violin Hauschka plays the instrument', 'Grant Hill is a professional soccer Magic Johnson is a professional', 'The law in Ikaalinen declares the language Swedish In Loviisa, the language spoken is'],
'ground_truth': ['piano', 'basketball', 'Finnish']
}
}
In the above example, we evaluate the performance of the editing methods about "neighborhood" and "distracting".
Step6: Edit and Evaluation
Done! We can conduct Edit and Evaluation for your model to be edited. The edit
function will return a series of metrics related to the editing process as well as the modified model weights.
metrics, edited_model, _ = editor.edit(
prompts=prompts,
ground_truth=ground_truth,
target_new=target_new,
locality_inputs=locality_inputs,
keep_original_weight=False
)
## metrics: edit success, rephrase success, locality e.g.
## edited_model: post-edit model
We specify the return metrics as dict
format, including model prediction evaluations before and after editing. For each edit, it will include the following metrics:
-
rewrite_acc
$\rightarrow$ Reliablilty -
rephrase_acc
$\rightarrow$ Generalization -
locality
$\rightarrow$ Locality -
portablility
$\rightarrow$ Portablility
{
"post": {
"rewrite_acc": ,
"rephrase_acc": ,
"locality": {
"YOUR_LOCALITY_KEY": ,
//...
},
"portablility": {
"YOUR_PORTABILITY_KEY": ,
//...
},
},
"pre": {
"rewrite_acc": ,
"rephrase_acc": ,
"portablility": {
"YOUR_PORTABILITY_KEY": ,
//...
},
}
}
- For evaluation for Reliablilty, you only need to provide the corresponding editing
prompts
and editingtarget_new
. - For evaluation for Generalization,
rephrase_prompts
are required. - For evaluation for Locality and Portablility, you need to define the name of the corresponding metric, as well as
prompts
andground_truth
.-
Note: the length needs to be equal to the edit prompts
-
- meta-learning based:
MEND
- memory-based routing:
SERAC
For above editing methods, pre-training of corresponding meta-networks or classifiers is required. Therefore, in EasyEdit, we provide a unified framework for pretraining the relevant network structures. Take the training MEND for example:
- Step 1 and Step 2 are the same as the example above, which involves selecting the appropriate editing model and editing method.
Step3: Provide the edit training set
The currently supported and available datasets are: zsre
and counterfact
(Google Drive). Please place them in the "data" directory and initialize the dataset_class (ZsreDataset
for zsre and CounterFactDataset
for counterfact) to load the corresponding training set.
train_ds = ZsreDataset('./data/zsre_mend_train.json', config=training_hparams)
eval_ds = ZsreDataset('./data/zsre_mend_eval.json', config=training_hparams)
Step4: Combine them into a Trainer
trainer = EditTrainer(
config=training_hparams,
train_set=train_ds,
val_set=eval_ds
)
Step5: Run and Edit Done! We can conduct Run and Evaluation.
trainer.run()
- Run: The
CHECKPOINT
will be saved to the pathresults_dir
. - Edit: Set the
archive
field in the hparams file toCHECKPOINT
. EasyEdit will automatically load the corresponding pre-trained weights during the editing process(Go to edit).
Training Example
from easyeditor import EditTrainer, MENDTrainingHparams, ZsreDataset
training_hparams = MENDTrainingHparams.from_hparams('hparams/TRAINING/MEND/llama-7b.yaml')
train_ds = ZsreDataset('./data/zsre/zsre_mend_train.json', config=training_hparams)
eval_ds = ZsreDataset('./data/zsre/zsre_mend_eval.json', config=training_hparams)
trainer = EditTrainer(
config=training_hparams,
train_set=train_ds,
val_set=eval_ds
)
trainer.run()
MultimodalEditor
is the class for Multi-Modality Editing. You can choose the appropriate editing method based on your specific needs.
- Due to different transformer versions and different GPU models, the editing results may fluctuate slightly.
M-Generality Results
VQA | KE | IKE | SERAC | MEND |
---|---|---|---|---|
MiniGPT-4 | 88.60 | 99.95 | 88.10 | 99.60 |
BLIP2 | 74.60 | 99.79 | 99.20 | 99.40 |
Caption | KE | IKE | SERAC | MEND |
---|---|---|---|---|
MiniGPT-4 | 13.60 | 91.00 | 91.47 | 93.35 |
BLIP2 | 1.60 | 96.55 | 99.72 | 93.48 |
With the modularity and flexibility of EasyEdit
, you can easily use it to edit model.
Step1: Define a MLLM as the object to be edited.
Choose the MLLM to be edited. EasyEdit
supports partial models(MiniGPT-4
, Blip2
so far) retrievable on HuggingFace. The corresponding configuration file directory is hparams/YUOR_METHOD/YOUR_MODEL.YAML
, such as hparams/MEND/minigpt4.yaml
, set the corresponding model_name
to select the object for editing.
model_name: minigpt4
model_class: Blip2OPT
tokenizer_class: LlamaTokenizer
tokenizer_name: llama-7b
Step2: Choose the appropriate Editing Method The selection of editing methods is a crucial step, as different methods have their own strengths and weaknesses. Users need to consider the trade-off between editing success rate, generalization, and maintaining unrelated performance.
## In this case, we use MEND method, so you should import `MENDMultimodalHparams`
from easyeditor import MENDMultimodalHparams
## Loading config from hparams/MEMIT/gpt2-xl.yaml
hparams = MENDMultimodalHparams.from_hparams('./hparams/MEND/minigpt4')
Step3: Provide the edit descriptor and edit target
## edit descriptor: prompt that you want to edit
prompts = [
"How many tennis balls are in the picture?",
"What is the red food?"
]
## edit target: expected output
targets = ["2", "tomatoes",]
## edit image: image for editing
image = [
"val2014/COCO_val2014_000000451435.jpg",
"val2014/COCO_val2014_000000189446.jpg"
]
Step4: Combine them into a MultimodalEditor
EasyEdit
provides a simple and unified way to init Editor, like huggingface: from_hparams.
## Construct MLLM Editor
editor = MultimodalEditor.from_hparams(hparams)
Step5: Provide the data for evaluation Note that the data for locality and multimodal locality are both optional(set to None for basic editing success rate evaluation only). The data format for both is a dict, for each measurement dimension, you need to provide the corresponding prompt and its corresponding ground truth. Here is an example of the data:
locality_inputs = {
'text': {
'prompt': [
"nq question: what purpose did seasonal monsoon winds have on trade"
],
'ground_truth': [
"enabled European empire expansion into the Americas and trade \
routes to become established across the Atlantic and Pacific oceans"
]
},
'vision': {
'prompt': ["What sport can you use this for?"],
'ground_truth': ["riding"],
'image': ["val2014/COCO_val2014_000000297147.jpg"],
}
}
In the above example, we evaluate the performance of the editing methods about "neighborhood" and "distracting".
Step6: Edit and Evaluation
Done! We can conduct Edit and Evaluation for your model to be edited. The edit
function will return a series of metrics related to the editing process as well as the modified model weights.
metrics, edited_model, _ = editor.edit(
prompts=prompts,
target_new=target_new,
image=image,
locality_inputs=locality_inputs,
keep_original_weight=False
)
## metrics: edit success, rephrase success, locality e.g.
## edited_model: post-edit model
We specify the return metrics as dict
format, including model prediction evaluations before and after editing. For each edit, it will include the following metrics:
-
rewrite_acc
$\rightarrow$ Reliablilty -
rephrase_acc
$\rightarrow$ Generalization -
image_rephrase_acc
$\rightarrow$ Generalization for Multimodal -
locality_acc
$\rightarrow$ Locality -
multimodal_locality_acc
$\rightarrow$ Locality for Multimodal
{
"post": {
"rewrite_acc": ,
"rephrase_acc": ,
"image_rephrase_acc": ,
"locality_acc": ,
"multimodal_locality_acc": ,
},
"pre": {
"rewrite_acc": ,
"rephrase_acc": ,
"image_rephrase_acc": ,
}
}
- For evaluation for Reliablilty, you only need to provide the corresponding editing
prompts
and editingtarget_new
. - For evaluation for Generalization,
rephrase_prompts
are required. - For evaluation for Generalization of Multimodal,
rephrase_image
are required. - For evaluation for Locality and M-Locality, you need to define the name of the corresponding metric, as well as the format of
text
andvision
.-
Note: the length needs to be equal to the edit prompts
-
- meta-learning based:
MEND
- memory-based routing:
SERAC
For above editing methods, pre-training of corresponding meta-networks or classifiers is required. Therefore, in EasyEdit, we provide a unified framework for pretraining the relevant network structures. Take the training SERAC for example:
- Step 1 and Step 2 are the same as the example above, which involves selecting the appropriate editing model and editing method.
Step3: Provide the edit training set
The currently supported and available datasets are: Caption
and VQA
(Google Drive). Please place them in the "data" directory and initialize the dataset_class (CaptionDataset
for Caption and VQADataset
for VQA) to load the corresponding training set.
train_ds = CaptionDataset('data/caption_train_edit.json', config=training_hparams)
eval_ds = CaptionDataset('data/caption_eval_edit.json', config=training_hparams)
Step4: Combine them into a Trainer
trainer = MultimodalTrainer(
config=hparams,
train_set=train_ds,
val_set=eval_ds
)
Step5: Run and Edit Done! We can conduct Run and Evaluation.
trainer.run()
- Run: The
CHECKPOINT
will be saved to the pathresults_dir
. - Edit: Set the
archive
field in the hparams file toCHECKPOINT
. EasyEdit will automatically load the corresponding pre-trained weights during the editing process(Go to edit).
Training Example
hparams = SERACMultimodalTrainingHparams.from_hparams('hparams/TRAINING/SERAC/minigpt4.yaml')
train_ds = CaptionDataset('data/caption_train_edit.json', config=training_hparams)
eval_ds = CaptionDataset('data/caption_eval_edit.json', config=training_hparams)
trainer = MultimodalTrainer(
config=hparams,
train_set=train_ds,
val_set=eval_ds
)
trainer.run()
TO DO
In next version, we plan to:- Explore and integrate more robust editing methods, focusing on
locality
andportability
metrics. - Provide a comprehensive evaluation suite for editing methods, including fact modification, fact erasure and hallucination erasure.
- Provide a causal analysis component for analyzing knowledge storage mechanisms.
- knowledge editing for other tasks(except factual editing), like
personality editing
, etc.
Meanwhile, we will offer long-term maintenance to fix bugs, solve issues and meet new requests. So if you have any problems, please put issues to us.
KnowEdit is a benchmark dataset of knowledge editing for LLMs. You can easily obtain KnowEdit from HuggingFace, HuggingFace, and ModelScope.
dataset | HuggingFace | HuggingFace | ModelScope |
---|---|---|---|
KnowEdit | [HuggingFace] | [WiseModel] | [ModelScope] |
We provide detailed scripts for user to easily use KnowEdit, please refer to examples.
We present editing results of the four metrics on LlaMA-2-7B using EasyEdit. We adopt ZsRE as the test dataset.
❗️❗️Editing
llama-2-7B
requires 40G+ VRAM on GPU. (OOM solution)
Reliability | Generalization | Locality | Portability | |
---|---|---|---|---|
FT | 56.94 | 52.02 | 96.32 | 0.07 |
SERAC | 99.49 | 99.13 | 100.00 | 0.13 |
IKE | 100.00 | 99.98 | 69.19 | 67.56 |
MEND | 94.24 | 90.27 | 97.04 | 0.14 |
KN | 28.95 | 28.43 | 65.43 | 0.07 |
ROME | 92.45 | 87.04 | 99.63 | 10.46 |
MEMIT | 92.94 | 85.97 | 99.49 | 6.03 |
We also present editing results of KnowEdit on LlaMA-2-7B using EasyEdit.
DataSet | Metric | SERAC | ICE | AdaLoRA | MEND | ROME | MEMIT | FT-L | FT |
---|---|---|---|---|---|---|---|---|---|
WikiData_recent | |||||||||
Edit Succ. | 98.68 | 60.74 | 65.61 | 76.88 | 85.08 | 85.32 | 71.18 | 31.24 | |
Portability | 63.52 | 36.93 | 47.22 | 50.11 | 37.45 | 37.94 | 48.71 | 15.91 | |
Locality | 100.00 | 33.34 | 55.78 | 92.87 | 66.2 | 64.78 | 63.7 | 3.65 | |
Fluency | 553.19 | 531.01 | 537.51 | 586.34 | 574.28 | 566.66 | 549.35 | 428.67 | |
ZsRE | |||||||||
Edit Succ. | 99.67 | 66.01 | 69.86 | 96.74 | 96.57 | 83.07 | 54.65 | 36.88 | |
Portability | 56.48 | 63.94 | 52.95 | 60.41 | 52.20 | 51.43 | 45.02 | 8.72 | |
Locality | 30.23 | 23.14 | 72.21 | 92.79 | 27.14 | 25.46 | 71.12 | 0.31 | |
Fluency | 410.89 | 541.14 | 532.82 | 524.33 | 570.47 | 559.72 | 474.18 | 471.29 | |
WikiBio | |||||||||
Edit Succ. | 99.69 | 95.53 | 97.02 | 93.66 | 95.05 | 94.29 | 66.27 | 95.64 | |
Locality | 69.79 | 47.90 | 57.87 | 69.51 | 46.96 | 51.56 | 60.14 | 13.38 | |
Fluency | 606.95 | 632.92 | 615.86 | 609.39 | 617.25 | 616.65 | 604.00 | 589.22 | |
WikiData_counterfact | |||||||||
Edit Succ. | 99.99 | 69.83 | 72.14 | 78.82 | 83.21 | 83.41 | 51.12 | 26.78 | |
Portability | 76.07 | 45.32 | 55.17 | 57.53 | 38.69 | 40.09 | 39.07 | 16.94 | |
Locality | 98.96 | 32.38 | 66.78 | 94.16 | 65.4 | 63.68 | 62.51 | 0.29 | |
Fluency | 549.91 | 547.22 | 553.85 | 588.94 | 578.84 | 568.58 | 544.80 | 483.71 | |
ConvSent | |||||||||
Edit Succ. | 62.75 | 52.78 | 44.89 | 50.76 | 45.79 | 44.75 | 49.50 | 61.93 | |
Locality | 0.26 | 49.73 | 0.18 | 3.42 | 0.00 | 0.00 | 0.00 | 0.00 | |
Fluency | 458.21 | 621.45 | 606.42 | 379.43 | 606.32 | 602.62 | 607.86 | 546.24 | |
Sanitation | |||||||||
Edit Succ. | 0.00 | 72.50 | 2.50 | 0.00 | 85.00 | 48.75 | 0.00 | 60.00 | |
Locality | 100.00 | 56.58 | 65.50 | 5.29 | 50.31 | 67.47 | 14.78 | 42.61 | |
Fluency | 416.29 | 794.15 | 330.44 | 407.18 | 465.12 | 466.10 | 439.10 | 351.39 |
Please cite our paper if you use EasyEdit in your work.
@article{zhang2024comprehensive,
title={A Comprehensive Study of Knowledge Editing for Large Language Models},
author={Zhang, Ningyu and Yao, Yunzhi and Tian, Bozhong and Wang, Peng and Deng, Shumin and Wang, Mengru and Xi, Zekun and Mao, Shengyu and Zhang, Jintian and Ni, Yuansheng and others},
journal={arXiv preprint arXiv:2401.01286},
year={2024}
}
@article{wang2023easyedit,
title={Easyedit: An easy-to-use knowledge editing framework for large language models},
author={Wang, Peng and Zhang, Ningyu and Xie, Xin and Yao, Yunzhi and Tian, Bozhong and Wang, Mengru and Xi, Zekun and Cheng, Siyuan and Liu, Kangwei and Zheng, Guozhou and others},
journal={arXiv preprint arXiv:2308.07269},
year={2023}
}
@article{yao2023editing,
title={Editing Large Language Models: Problems, Methods, and Opportunities},
author={Yao, Yunzhi and Wang, Peng and Tian, Bozhong and Cheng, Siyuan and Li, Zhoubo and Deng, Shumin and Chen, Huajun and Zhang, Ningyu},
journal={arXiv preprint arXiv:2305.13172},
year={2023}
}
@article{cheng2023edit,
title={Can We Edit Multimodal Large Language Models?},
author={Cheng, Siyuan and Tian, Bozhong and Liu, Qingbin and Chen, Xi and Wang, Yongheng and Chen, Huajun and Zhang, Ningyu},
journal={arXiv preprint arXiv:2310.08475},
year={2023}
}
@article{mao2023editing,
title={Editing personality for llms},
author={Mao, Shengyu and Zhang, Ningyu and Wang, Xiaohan and Wang, Mengru and Yao, Yunzhi and Jiang, Yong and Xie, Pengjun and Huang, Fei and Chen, Huajun},
journal={arXiv preprint arXiv:2310.02168},
year={2023}
}
@misc{knowlm,
author = {Ningyu Zhang and Jintian Zhang and Xiaohan Wang and Honghao Gui and Kangwei Liu and Yinuo Jiang and Xiang Chen and Shengyu Mao and Shuofei Qiao and Yuqi Zhu and Zhen Bi and Jing Chen and Xiaozhuan Liang and Yixin Ou and Runnan Fang and Zekun Xi and Xin Xu and Lei Li and Peng Wang and Mengru Wang and Yunzhi Yao and Bozhong Tian and Yin Fang and Guozhou Zheng and Huajun Chen},
title = {KnowLM Technical Report},
year = {2023},
url = {http://knowlm.zjukg.cn/},
}
We thank all the contributors to this project, more contributors are welcome!
🙌 We would like to express our heartfelt gratitude for the contribution of FastEdit, ROME, GRACE, MELO, PMET to our project, as we have utilized portions of their source code in our project. Many thanks to all the colleagues in the community for submitting issues and providing technical support.