diff --git a/modules/model_quant.py b/modules/model_quant.py index 31dcedbf0..686703b9b 100644 --- a/modules/model_quant.py +++ b/modules/model_quant.py @@ -428,9 +428,9 @@ def sdnq_quantize_model(model, op=None, sd_model=None, do_gc: bool = True, weigh non_blocking=shared.opts.diffusers_offload_nonblocking, quantization_device=quantization_device, return_device=return_device, - param_name=op, modules_to_not_convert=modules_to_not_convert, modules_dtype_dict=modules_dtype_dict, + op=op, ) t1 = time.time() timer.load.add('sdnq', t1 - t0) diff --git a/modules/sdnq/__init__.py b/modules/sdnq/__init__.py index ba4f7b9fb..2654a06e7 100644 --- a/modules/sdnq/__init__.py +++ b/modules/sdnq/__init__.py @@ -186,7 +186,7 @@ def sdnq_quantize_layer(layer, weights_dtype="int8", torch_dtype=None, group_siz return layer -def apply_sdnq_to_module(model, weights_dtype="int8", torch_dtype=None, group_size=0, quant_conv=False, use_quantized_matmul=False, use_quantized_matmul_conv=False, dequantize_fp32=False, non_blocking=False, quantization_device=None, return_device=None, param_name=None, modules_to_not_convert: List[str] = [], modules_dtype_dict: Dict[str, List[str]] = {}): # pylint: disable=unused-argument +def apply_sdnq_to_module(model, weights_dtype="int8", torch_dtype=None, group_size=0, quant_conv=False, use_quantized_matmul=False, use_quantized_matmul_conv=False, dequantize_fp32=False, non_blocking=False, quantization_device=None, return_device=None, modules_to_not_convert: List[str] = [], modules_dtype_dict: Dict[str, List[str]] = {}, op=None): # pylint: disable=unused-argument has_children = list(model.children()) if not has_children: return model @@ -238,9 +238,9 @@ def apply_sdnq_to_module(model, weights_dtype="int8", torch_dtype=None, group_si non_blocking=non_blocking, quantization_device=quantization_device, return_device=return_device, - param_name=param_name, modules_to_not_convert=modules_to_not_convert, modules_dtype_dict=modules_dtype_dict, + op=op, ) return model