Improve vanilla node importing and fix CUDA on CPU devices bug

comfy/component_model/plugins.py

@@ -6,7 +6,7 @@ class _RoutesWrapper:
         def decorator(path):
             def wrapper(func):
                 from ..cmd.server import PromptServer
-                if PromptServer.instance is not None:
+                if PromptServer.instance is not None and not isinstance(PromptServer.instance.routes, _RoutesWrapper):
                     getattr(PromptServer.instance.routes, method)(path)(func)
                 self.routes.append((method, path, func))
                 return func

comfy/model_management.py

@@ -42,6 +42,7 @@ model_management_lock = RLock()
 # This setting optimizes performance on NVIDIA GPUs with Ampere architecture (e.g., A100, RTX 30 series) or newer.
 torch.set_float32_matmul_precision("high")
 
+
 class VRAMState(Enum):
     DISABLED = 0  # No vram present: no need to move models to vram
     NO_VRAM = 1  # Very low vram: enable all the options to save vram
@@ -978,10 +979,12 @@ def cast_to_device(tensor, device, dtype, copy=False):
         else:
             return tensor.to(device, dtype, copy=copy, non_blocking=non_blocking)
 
+
 FLASH_ATTENTION_ENABLED = False
 if not args.disable_flash_attn:
     try:
         import flash_attn
+
         FLASH_ATTENTION_ENABLED = True
     except ImportError:
         pass
@@ -990,6 +993,7 @@ SAGE_ATTENTION_ENABLED = False
 if not args.disable_sage_attention:
     try:
         import sageattention
+
         SAGE_ATTENTION_ENABLED = True
     except ImportError:
         pass
@@ -1006,6 +1010,7 @@ def xformers_enabled():
         return False
     return XFORMERS_IS_AVAILABLE
 
+
 def flash_attn_enabled():
     global directml_device
     global cpu_state
@@ -1017,6 +1022,7 @@ def flash_attn_enabled():
         return False
     return FLASH_ATTENTION_ENABLED
 
+
 def sage_attention_enabled():
     global directml_device
     global cpu_state
@@ -1250,7 +1256,11 @@ def supports_fp8_compute(device=None):
     if not is_nvidia():
         return False
 
-    props = torch.cuda.get_device_properties(device)
+    try:
+        props = torch.cuda.get_device_properties(device)
+    except (RuntimeError, ValueError, AssertionError):
+        return False
+
     if props.major >= 9:
         return True
     if props.major < 8:

comfy/nodes/vanilla_node_importing.py

@@ -13,6 +13,12 @@ from os.path import join, basename, dirname, isdir, isfile, exists, abspath, spl
 
 from . import base_nodes
 from .package_typing import ExportedNodes
+from ..component_model.plugins import prompt_server_instance_routes
+
+
+class _PromptServerStub():
+    def __init__(self):
+        self.routes = prompt_server_instance_routes
 
 
 def _vanilla_load_importing_execute_prestartup_script(node_paths: Iterable[str]) -> None:
@@ -176,6 +182,9 @@ def mitigated_import_of_vanilla_custom_nodes() -> ExportedNodes:
         module_short_name = module.__name__.split(".")[-1]
         sys.modules[module_short_name] = module
 
+    if server.PromptServer.instance is None:
+        server.PromptServer.instance = _PromptServerStub()
+
     # Impact Pack wants to find model_patcher
     from .. import model_patcher
     sys.modules['model_patcher'] = model_patcher