From d39e7f9e68f41bd9a6ec53858ab21bf0c7480f99 Mon Sep 17 00:00:00 2001 From: George Ohashi Date: Mon, 17 Jun 2024 19:58:29 +0000 Subject: [PATCH] comments --- src/sparseml/modifiers/quantization/gptq/utils/gptq_wrapper.py | 1 + src/sparseml/modifiers/utils/layer_compressor.py | 2 ++ 2 files changed, 3 insertions(+) diff --git a/src/sparseml/modifiers/quantization/gptq/utils/gptq_wrapper.py b/src/sparseml/modifiers/quantization/gptq/utils/gptq_wrapper.py index de472c66ecb..30a13196d92 100644 --- a/src/sparseml/modifiers/quantization/gptq/utils/gptq_wrapper.py +++ b/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 diff --git a/src/sparseml/modifiers/utils/layer_compressor.py b/src/sparseml/modifiers/utils/layer_compressor.py index 5090539d84e..2d7fdf53e00 100644 --- a/src/sparseml/modifiers/utils/layer_compressor.py +++ b/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()