EasyEdit issues

Recommended ROME layers and v_loss_layer for Qwen3 series (0.6B, 1.7B, 4B, 8B)

4

Hi EasyEdit Team, First, thank you for this fantastic library! I am currently working on adding support for the new Qwen3 model series (0.6B, 1.7B, 4B, and 8B) to use...

Salehoof

question

The metric summary for pre and post is the same when using UltraEdit

5

Thanks for your repo. When I used UltraEdit to edit Qwen2.5, I found that the metric summary was the same for pre and post. Is this normal? ` Metrics Summary:...

chenyangzhu1

question

Build-in validation in MultimodalTrainer and evaluation in MultimodalEditor

1

I notice that there is a validate() function in `MultimodalTrainer` and you validate the trained model after training. Since the validate() function edits the model with the validation set and...

Wang-Z-W

question

Output of gen_text

When I run the WISE editing method with a LLaMA configuration (e.g., using run_wise_editing.sh), the example code only produces accuracy/evaluation metrics. I expected to see the generated texts (gen_text) as...

Tuner12

question

cache_c accumulation in AlphaEdit during single-edit testing

Hi, thank you for sharing this great work! While testing AlphaEdit with independent single edits, I noticed that the global variable cache_c seems to keep accumulating across calls, even though...

dannalily

question

GRACE's performance collapes on LLAMA-3-8B

4

Thanks a lot for the great works, I want to ask about the performance of GRACE with LLAMA-3-8B model. My reproduced results with GRACE, as well as the results reported...

wang-kee

question

5090 compatibility

I hope this message finds you well. Would it be possible to kindly let me know if you have conducted any experiments using the 5090 GPUs? If so, I would...

Wang-Z-W

question

SERAC Applied on Other Datasets

Hi, I tried to apply SERAC to our constructed datasets in MLLM, but the performance turned out to be very low, with a rewrite accuracy of around 0.1. It tends...

Hz7even