Fix dora.

comfy/lora.py

@@ -327,6 +327,26 @@ def model_lora_keys_unet(model, key_map={}):
     return key_map
 
 
+def weight_decompose(dora_scale, weight, lora_diff, alpha, strength, intermediate_dtype):
+    dora_scale = comfy.model_management.cast_to_device(dora_scale, weight.device, intermediate_dtype)
+    lora_diff *= alpha
+    weight_calc = weight + lora_diff.type(weight.dtype)
+    weight_norm = (
+        weight_calc.transpose(0, 1)
+        .reshape(weight_calc.shape[1], -1)
+        .norm(dim=1, keepdim=True)
+        .reshape(weight_calc.shape[1], *[1] * (weight_calc.dim() - 1))
+        .transpose(0, 1)
+    )
+
+    weight_calc *= (dora_scale / weight_norm).type(weight.dtype)
+    if strength != 1.0:
+        weight_calc -= weight
+        weight += strength * (weight_calc)
+    else:
+        weight[:] = weight_calc
+    return weight
+
 def calculate_weight(patches, weight, key, intermediate_dtype=torch.float32):
     for p in patches:
         strength = p[0]

comfy/model_detection.py

@@ -472,9 +472,15 @@ def unet_config_from_diffusers_unet(state_dict, dtype=None):
             'transformer_depth': [0, 1, 1], 'channel_mult': [1, 2, 4], 'transformer_depth_middle': -2, 'use_linear_in_transformer': False,
             'context_dim': 768, 'num_head_channels': 64, 'transformer_depth_output': [0, 0, 1, 1, 1, 1],
             'use_temporal_attention': False, 'use_temporal_resblock': False}
+    
+    SD15_diffusers_inpaint = {'use_checkpoint': False, 'image_size': 32, 'out_channels': 4, 'use_spatial_transformer': True, 'legacy': False, 'adm_in_channels': None,
+            'dtype': dtype, 'in_channels': 9, 'model_channels': 320, 'num_res_blocks': [2, 2, 2, 2], 'transformer_depth': [1, 1, 1, 1, 1, 1, 0, 0],
+            'channel_mult': [1, 2, 4, 4], 'transformer_depth_middle': 1, 'use_linear_in_transformer': False, 'context_dim': 768, 'num_heads': 8,
+            'transformer_depth_output': [1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0],
+            'use_temporal_attention': False, 'use_temporal_resblock': False}  
 
 
-    supported_models = [SDXL, SDXL_refiner, SD21, SD15, SD21_uncliph, SD21_unclipl, SDXL_mid_cnet, SDXL_small_cnet, SDXL_diffusers_inpaint, SSD_1B, Segmind_Vega, KOALA_700M, KOALA_1B, SD09_XS, SD_XS, SDXL_diffusers_ip2p]
+    supported_models = [SDXL, SDXL_refiner, SD21, SD15, SD21_uncliph, SD21_unclipl, SDXL_mid_cnet, SDXL_small_cnet, SDXL_diffusers_inpaint, SSD_1B, Segmind_Vega, KOALA_700M, KOALA_1B, SD09_XS, SD_XS, SDXL_diffusers_ip2p, SD15_diffusers_inpaint]
 
     for unet_config in supported_models:
         matches = True

comfy/model_management.py

@@ -44,9 +44,14 @@ cpu_state = CPUState.GPU
 
 total_vram = 0
 
-lowvram_available = True
 xpu_available = False
+try:
+    torch_version = torch.version.__version__
+    xpu_available = (int(torch_version[0]) < 2 or (int(torch_version[0]) == 2 and int(torch_version[2]) <= 4)) and torch.xpu.is_available()
+except:
+    pass
 
+lowvram_available = True
 if args.deterministic:
     logging.info("Using deterministic algorithms for pytorch")
     torch.use_deterministic_algorithms(True, warn_only=True)
@@ -66,10 +71,10 @@ if args.directml is not None:
 
 try:
     import intel_extension_for_pytorch as ipex
-    if torch.xpu.is_available():
-        xpu_available = True
+    _ = torch.xpu.device_count()
+    xpu_available = torch.xpu.is_available()
 except:
-    pass
+    xpu_available = xpu_available or (hasattr(torch, "xpu") and torch.xpu.is_available())
 
 try:
     if torch.backends.mps.is_available():
@@ -189,7 +194,6 @@ VAE_DTYPES = [torch.float32]
 
 try:
     if is_nvidia():
-        torch_version = torch.version.__version__
         if int(torch_version[0]) >= 2:
             if ENABLE_PYTORCH_ATTENTION == False and args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
                 ENABLE_PYTORCH_ATTENTION = True
@@ -321,8 +325,9 @@ class LoadedModel:
                 self.model_unload()
                 raise e
 
-        if is_intel_xpu() and not args.disable_ipex_optimize:
-            self.real_model = ipex.optimize(self.real_model.eval(), graph_mode=True, concat_linear=True)
+        if is_intel_xpu() and not args.disable_ipex_optimize and self.real_model is not None:
+            with torch.no_grad():
+                self.real_model = ipex.optimize(self.real_model.eval(), inplace=True, graph_mode=True, concat_linear=True)
 
         self.weights_loaded = True
         return self.real_model

comfy/model_patcher.py

@@ -31,27 +31,6 @@ import comfy.lora
 from comfy.types import UnetWrapperFunction
 
 
-def weight_decompose(dora_scale, weight, lora_diff, alpha, strength, intermediate_dtype):
-    dora_scale = comfy.model_management.cast_to_device(dora_scale, weight.device, intermediate_dtype)
-    lora_diff *= alpha
-    weight_calc = weight + lora_diff.type(weight.dtype)
-    weight_norm = (
-        weight_calc.transpose(0, 1)
-        .reshape(weight_calc.shape[1], -1)
-        .norm(dim=1, keepdim=True)
-        .reshape(weight_calc.shape[1], *[1] * (weight_calc.dim() - 1))
-        .transpose(0, 1)
-    )
-
-    weight_calc *= (dora_scale / weight_norm).type(weight.dtype)
-    if strength != 1.0:
-        weight_calc -= weight
-        weight += strength * (weight_calc)
-    else:
-        weight[:] = weight_calc
-    return weight
-
-
 def set_model_options_patch_replace(model_options, patch, name, block_name, number, transformer_index=None):
     to = model_options["transformer_options"].copy()