From 4e3795a0a59225bac7ff13b98a3a1a9dd6094d0a Mon Sep 17 00:00:00 2001 From: Disty0 Date: Mon, 2 Jun 2025 03:31:51 +0300 Subject: [PATCH] SDNQ fix packed int8 matmul --- modules/model_quant_sdnq.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/model_quant_sdnq.py b/modules/model_quant_sdnq.py index a4416860f..8231a3828 100644 --- a/modules/model_quant_sdnq.py +++ b/modules/model_quant_sdnq.py @@ -418,7 +418,7 @@ def int8_matmul( weights_dtype: str, ) -> torch.FloatTensor: if compressed_weight_shape is not None: - weight = packed_int_function_dict[weights_dtype]["unpack"](weight, compressed_weight_shape, transpose=True) + weight = unpack_int_symetric(weight, compressed_weight_shape, weights_dtype, dtype=torch.int8, transpose=True) return_dtype = input.dtype output_shape = list(input.shape) output_shape[-1] = weight.shape[-1]