comments

Author: horheynm

src/sparseml/modifiers/quantization/gptq/utils/gptq_wrapper.py

@@ -89,6 +89,7 @@ def fasterprune(
         Run pruning and quantization(if applicable) on the layer up to the target
         sparsity value.
 
+        :param actorder: Flag to apply activation reordering
         :param blocksize: Number of columns to compress in one pass
         :param percdamp: Amount of dampening to apply to H, as a fraction of the
             diagonal norm

src/sparseml/modifiers/utils/layer_compressor.py

@@ -134,6 +134,8 @@ def revert_layer_wrappers(self):
     def compress(self, actorder: bool = False):
         """
         Apply compression to each wrapped submodule in the layer
+        
+        :param: actorder: flag to apply activation reordering
         """
 
         @torch.no_grad()