Merge pull request #161 from intel/gpu_precision_fix

Fix SD1.5 FP16 for some ARL and MTL GPU SKUs.
intel · Jan 11, 2025 · 0846f9e · 0846f9e
2 parents 6c5284c + 23a1ce1
commit 0846f9e
Showing 1 changed file with 5 additions and 3 deletions.
diff --git a/...penvino/plugins/openvino_utils/tools/openvino_common/models_ov/stable_diffusion_engine.py b/...penvino/plugins/openvino_utils/tools/openvino_common/models_ov/stable_diffusion_engine.py
@@ -523,13 +523,15 @@ def __init__(
 
         self.set_dimensions()
 
-
-
     def load_model(self, model, model_name, device):
         if "NPU" in device:
             with open(os.path.join(model, f"{model_name}.blob"), "rb") as f:
                 return self.core.import_model(f.read(), device)
-        return self.core.compile_model(os.path.join(model, f"{model_name}.xml"), device)
+        if "GPU" in device:
+            return self.core.compile_model(os.path.join(model, f"{model_name}.xml"), device, {'INFERENCE_PRECISION_HINT': 'f32'})
+        else:
+            return self.core.compile_model(os.path.join(model, f"{model_name}.xml"), device)
+
 
     def set_dimensions(self):
         latent_shape = self.unet.input(self.unet_input_tensor_name).shape