diff --git a/gptqmodel/nn_modules/qlinear/qlinear_exllama.py b/gptqmodel/nn_modules/qlinear/qlinear_exllama.py index 40341b26..bb4e4cde 100644 --- a/gptqmodel/nn_modules/qlinear/qlinear_exllama.py +++ b/gptqmodel/nn_modules/qlinear/qlinear_exllama.py @@ -55,15 +55,9 @@ def __init__(self, bits: int, group_size: int , sym:bool, desc_act: bool, infeat self.original_outfeatures = outfeatures self.original_infeatures = infeatures - # code bug prevention - del infeatures - del outfeatures - del group_size - self.maxq = 2**self.bits - 1 assert self.infeatures % 32 == 0 - assert self.infeatures % self.group_size == 0 assert self.outfeatures % 32 == 0 self.register_buffer( diff --git a/gptqmodel/nn_modules/qlinear/qlinear_exllamav2.py b/gptqmodel/nn_modules/qlinear/qlinear_exllamav2.py index c387df8c..50506f9a 100644 --- a/gptqmodel/nn_modules/qlinear/qlinear_exllamav2.py +++ b/gptqmodel/nn_modules/qlinear/qlinear_exllamav2.py @@ -118,16 +118,9 @@ def __init__(self, bits: int, group_size: int, sym: bool, desc_act: bool, infeat # backup original values self.original_outfeatures = outfeatures self.original_infeatures = infeatures - - # code bug prevention - del infeatures - del outfeatures - del group_size - self.maxq = 2**self.bits - 1 assert self.infeatures % 32 == 0 - assert self.infeatures % self.group_size == 0 assert self.outfeatures % 32 == 0 # I need to register the tensors, otherwise, we won't be able to load them easily using transformers ...