Dear authors,
I would like to express my deep appreciation for the significant contribution your work has made to our community. It has been an invaluable resource for me.
I have encountered a question regarding the calibration set you chose when trying to figure out the code. I hope you can offer some suggestions.
In your implementation, when pruning LLM with FLAP, the calibration set is chosen as WikiText-v2, as shown in this line.
If I understand correctly, comparing the PPL on WikiText-v2 with previous work might be misleading, as previous pruning methods were not trained on WikiText-v2. Could you please confirm this?
Can you specify the calibration set you choose when you report the result in your paper? I can not find this in the paper.
Thanks for your attention, and I hope to hear from you!
Dear authors,
I would like to express my deep appreciation for the significant contribution your work has made to our community. It has been an invaluable resource for me.
I have encountered a question regarding the calibration set you chose when trying to figure out the code. I hope you can offer some suggestions.
In your implementation, when pruning LLM with FLAP, the calibration set is chosen as WikiText-v2, as shown in this line.
If I understand correctly, comparing the PPL on WikiText-v2 with previous work might be misleading, as previous pruning methods were not trained on WikiText-v2. Could you please confirm this?
Can you specify the calibration set you choose when you report the result in your paper? I can not find this in the paper.
Thanks for your attention, and I hope to hear from you!