EasyEdit
EasyEdit copied to clipboard
zjunlp
[ACL 2024] An Easy-to-use Knowledge Editing Framework for LLMs.
Hi! I'm trying to evaluate Mistral-7b based model with custom locality and portability data. For each of 50 edits I have 6 locality prompts and 2 portability ones. How should...
I used MEMIT to edit Qwen-14B model, running the demo adapted from the official example, and the result is pretty bad (shown as below). The edited model only generated exclamation...
Hi, thank you for this amazing work! I'm trying to edit `NousResearch/Llama-2-7b-chat-hf` using PMET. I'm following the samples you provided. However it gives exactly the same answer for both the...
Line 118 of `ft_main.py`: ``` loss_mask = target_ids ! = tok.unk_token_id ``` The handling of tok.unk_token_id as None seems to be omitted here. If tok.unk_token_id is None, then on line...
Hello! I came across a tiny issue while editing Llama2 using the PMET method, and I thought it might be worth mentioning: At line 44 of the file `easyeditor/models/pmet/compute_zs.py`: ```python...
Hi, Recent works found some instability when using ROME resulting in model collapse. We found that the original implementation of ROME was not the correct representation of update equation derived...
In function apply_ike_to_multimodal_model in EasyEdit-main/easyeditor/models/ike/ike_main.py, the last line is ``` icl_examples.append(f'New Fact: {new_fact}\nPrompt: {new_fact}\n\n') ``` I'm a little confused. Why adding the question and answer of a testing sample when...
I encountered some issues while using the KN method to edit the Llama2 and gpt2-xl models based on the easyedit framework. Here is the code snippet: editor = BaseEditor.from_hparams(hparams) metrics,...
您好，我在复现你们使用EasyEdit在LlaMA-2-7B上的指标的时候遇到了问题。如下图是你们在README中给出的测试结果  我尝试复现其中FT的结果，我使用命令 ``` python run_zsre_llama2.py \ --editing_method=FT \ --hparams_dir=../hparams/FT/llama-7b \ --data_dir=./data ``` llama-7b.yaml的内容是 ``` alg_name: "FT" model_name: "../hugging_cache/llama-2-7b" device: 0 layers: [21] num_steps: 25 batch_size: 1 max_length: 40 lr:...
MALMEN: Massive Language Model Editing Network [https://arxiv.org/abs/2311.04661](url) is the newest SOTA on Meta Learning method. In their experiments, it has the excellent performance with massive editing even better than MEMIT...
