From 3190e856de4df5193d0496accbe9c36ef7b24228 Mon Sep 17 00:00:00 2001 From: Disty0 Date: Tue, 13 Feb 2024 12:30:36 +0300 Subject: [PATCH] OpenVINO fix quant not being applied --- modules/intel/openvino/__init__.py | 16 ++++------------ 1 file changed, 4 insertions(+), 12 deletions(-) diff --git a/modules/intel/openvino/__init__.py b/modules/intel/openvino/__init__.py index 38b9fe35c..5023a2dcd 100644 --- a/modules/intel/openvino/__init__.py +++ b/modules/intel/openvino/__init__.py @@ -20,14 +20,6 @@ import functools from modules import shared, devices, sd_models -def get_node_by_name(self, name: str) -> nncf.common.graph.NNCFNode: - node_ids = self._node_name_to_node_id_map.get(name, None) - if node_ids is None: - raise RuntimeError("Could not find a node {} in NNCFGraph!".format(name)) - - node_key = f"{node_ids[0]} {name}" - return self._nodes[node_key] -nncf.common.graph.NNCFGraph.get_node_by_name = get_node_by_name def BUILD_MAP_UNPACK(self, inst): items = self.popn(inst.argval) @@ -242,9 +234,9 @@ def openvino_compile(gm: GraphModule, *example_inputs, model_hash_str: str = Non new_inputs.append(example_inputs[idx].detach().cpu().numpy()) new_inputs = [new_inputs] if shared.opts.nncf_quant_mode == "INT8": - nncf.quantize(om, nncf.Dataset(new_inputs)) + om = nncf.quantize(om, nncf.Dataset(new_inputs)) else: - nncf.quantize(om, nncf.Dataset(new_inputs), mode=getattr(nncf.QuantizationMode, shared.opts.nncf_quant_mode), + om = nncf.quantize(om, nncf.Dataset(new_inputs), mode=getattr(nncf.QuantizationMode, shared.opts.nncf_quant_mode), advanced_parameters=nncf.quantization.advanced_parameters.AdvancedQuantizationParameters( overflow_fix=nncf.quantization.advanced_parameters.OverflowFix.DISABLE, backend_params=None)) @@ -294,9 +286,9 @@ def openvino_compile_cached_model(cached_model_path, *example_inputs): new_inputs.append(example_inputs[idx].detach().cpu().numpy()) new_inputs = [new_inputs] if shared.opts.nncf_quant_mode == "INT8": - nncf.quantize(om, nncf.Dataset(new_inputs)) + om = nncf.quantize(om, nncf.Dataset(new_inputs)) else: - nncf.quantize(om, nncf.Dataset(new_inputs), mode=getattr(nncf.QuantizationMode, shared.opts.nncf_quant_mode), + om = nncf.quantize(om, nncf.Dataset(new_inputs), mode=getattr(nncf.QuantizationMode, shared.opts.nncf_quant_mode), advanced_parameters=nncf.quantization.advanced_parameters.AdvancedQuantizationParameters( overflow_fix=nncf.quantization.advanced_parameters.OverflowFix.DISABLE, backend_params=None))