ch_sparsity is 1.0 in main.py

[VincentMy]

What does it mean that this parameter is 1 in benchmarks/main.py “ch_sparsity” parameter should represent the removal rate， is that mean remove 100% channels ?

[VainF]

Hi @VincentMy, thanks for reaching out. In this example, the model undergoes progressive pruning, where the sparsity is increased from 0 to 1. This process stops once the desired speed-up is achieved.

[kewang-seu]

Is the increase of sparsity equal in step length？Is "iterative-steps" parameter the number of iteration？

[VainF]

Hi @kewang-seu, for a 10-step pruning with target sparsity=0.4, we will get somthing like:

0 -> 0.1 -> 0.2 -> 0.3 -> 0.4 (Terminal) -- 0.5 -- 0.6 -- ...