consult

[bo102]

Hello, I would like to ask how I can use your pruning method to prune the swin-t transformer-tiny model to get a model after pruning, I see that your search architecture part is used to search how to prune? Generate spvit_swin_t_l28_t32_search_12epoch.pth file to guide the next pruning? Or is it something else? Is that fine-tuning part used to restore the performance of the model after pruning? Thank you very much for your advice!``

[Charleshhy]

Hi bo102,

Thanks for your interest in our work! You are correct that:

1. spvit_swin_t_l28_t32_search_12epoch.pth is the output of the search process which guides the pruning process.
2. The fine-tuning part is used to restore the performance of the model after pruning.
3. Please see the configuration for pruning Swin-ti here and the fine-tuned weights without knowledge distillation here. You can also find more fine-tuned weights in this page.

Best, Haoyu

[bo102]

嗨，bo102，

感谢您对我们工作的关注！你是对的：

1. spvit_swin_t_l28_t32_search_12epoch.pth是指导修剪过程的搜索过程的输出。
2. 微调部分用于恢复修剪后模型的性能。
3. 请在此处查看修剪 Swin-ti 的配置，在此处查看无需知识蒸馏的微调砝码。您还可以在此页面中找到更多微调的权重。

贝斯特，浩宇

Can you send me a swin-transformer-tiny model file with the model structure and weights after pruning? (Only pruning, not knowledge distillation) github downloaded file. It doesn't seem to contain the structure of the model after pruning Thank you very much