AM69A: Not able to do pruning using edge-ai-tensorlab

My code for pruning - 

from edgeai_torchmodelopt.xmodelopt.pruning.v3 import create_channel_pruned_model
from edgeai_torchmodelopt import xmodelopt

####
####
####

model = xmodelopt.pruning.v3.PrunerModule(model, pruning_ratio=0.3, total_epochs=10)# edgeai-yolox model being loaded already

Error I am getting - 

  File "/workspace/edgeai-tensorlab/edgeai-modeloptimization/torchmodelopt/edgeai-yolox/check_prune.py", line 53, in <module>
    model = xmodelopt.pruning.v3.PrunerModule(model, pruning_ratio=0.3, total_epochs=10)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/workspace/edgeai-tensorlab/edgeai-modeloptimization/torchmodelopt/edgeai_torchmodelopt/xmodelopt/pruning/v3/pruner_module.py", line 55, in __init__
    self.prepare(module, *args, example_inputs= example_inputs, example_kwargs=example_kwargs, pruning_ratio=pruning_ratio, total_epochs=total_epochs, pruning_class=pruning_class, copy_args = copy_args,
  File "/workspace/edgeai-tensorlab/edgeai-modeloptimization/torchmodelopt/edgeai_torchmodelopt/xmodelopt/pruning/v3/pruner_module.py", line 60, in prepare
    copy_attrs = copy_attrs or []
                 ^^^^^^^^^^
UnboundLocalError: cannot access local variable 'copy_attrs' where it is not associated with a value

Hi Chris,

I have tried torch pruning. We wanted to try TI pruning as it is given in edgeai-tensorlab. 

Hi Ahkilesh,

I sent a message to our model curator.  Perhaps he knows of additional documentation.

Regards,

Chris

Hi Akhilesh, I have some questions:

Could you explain the goal of your activity? Do you want to do channel pruning or do you want to do weight sparsification such as 2:4 weight sparsity? On which SOC do you want to try to run the resultant model?

Hi Manu,

I am trying channel pruning for now and wanted to target am69a board. I have the code snippet and error while I am trying this.

Thanks

The error that you are facing can most probably be solved by replacing that line with error to 

copy_attrs = False

But, since the model structures widely vary, the channel pruning feature can fail when tried on a complex model. We have not been working on the pruning module for some time.

Hi Manu Mathew 

There were several error I encountered like that. Even if I resolve those, there is always some error. Seems like pruning is not working here. When can we expect this to be matured enough to use?

Thanks