From 26461f1d8d4d0ccac0dd3f1a7fd242263c23c0b1 Mon Sep 17 00:00:00 2001 From: Disty0 Date: Mon, 11 Aug 2025 23:15:25 +0300 Subject: [PATCH] fix conv in8 matmul --- modules/sdnq/layers/conv/conv_int8.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/sdnq/layers/conv/conv_int8.py b/modules/sdnq/layers/conv/conv_int8.py index c899116a6..1ebee0315 100644 --- a/modules/sdnq/layers/conv/conv_int8.py +++ b/modules/sdnq/layers/conv/conv_int8.py @@ -28,7 +28,7 @@ def conv_int8_matmul( input, mm_output_shape = process_conv_input(conv_type, input, reversed_padding_repeated_twice, padding_mode, result_shape, stride, padding, dilation) input, scale = quantize_int8_matmul_input(input, scale) if quantized_weight_shape is not None: - weight = unpack_int_symetric(weight, quantized_weight_shape, weights_dtype, dtype=torch.int8, transpose=True) + weight = unpack_int_symetric(weight, quantized_weight_shape, weights_dtype, dtype=torch.int8) if groups == 1: result = torch._int_mm(input, weight)