Improve tests, fix issues with alternate filenames, improve group offloading support for transformers models

README.md

@@ -242,7 +242,7 @@ ComfyUI LTS supports text and multi-modal LLM models from the `transformers` eco
 
 In this example, LLAVA-NEXT (LLAVA 1.6) is prompted to describe an image.
 
-You can try the [LLAVA-NEXT](tests/inference/workflows/llava-0.json), [Phi-3](tests/inference/workflows/phi-3-0.json), and two [translation](tests/inference/workflows/translation-0.json) [workflows](tests/inference/workflows/translation-1.json).
+You can try the [LLAVA-NEXT](tests/inference/workflows/llava-0.json), [Phi-3](tests/inference/workflows/phi-4-0.json), and two [translation](tests/inference/workflows/translation-0.json) [workflows](tests/inference/workflows/translation-1.json).
 
 # SVG Conversion and String Saving
 

comfy/cli_args.py

@@ -130,7 +130,7 @@ def _create_parser() -> EnhancedConfigArgParser:
     parser.add_argument("--force-non-blocking", action="store_true", help="Force ComfyUI to use non-blocking operations for all applicable tensors. This may improve performance on some non-Nvidia systems but can cause issues with some workflows.")
     parser.add_argument("--default-hashing-function", type=str, choices=['md5', 'sha1', 'sha256', 'sha512'], default='sha256', help="Allows you to choose the hash function to use for duplicate filename / contents comparison. Default is sha256.")
     parser.add_argument("--disable-smart-memory", action="store_true",
-                        help="Force ComfyUI to agressively offload to regular ram instead of keeping models in vram when it can.")
+                        help="Force ComfyUI to aggressively offload to regular ram instead of keeping models in VRAM when it can.")
     parser.add_argument("--deterministic", action="store_true",
                         help="Make pytorch use slower deterministic algorithms when it can. Note that this might not make images deterministic in all cases.")
 

comfy/language/transformers_model_management.py

@@ -12,9 +12,12 @@ from typing import Optional, Any, Callable
 
 import torch
 import transformers
+from huggingface_hub.errors import EntryNotFoundError
 from transformers import PreTrainedModel, PreTrainedTokenizerBase, ProcessorMixin, AutoProcessor, AutoTokenizer, \
     BatchFeature, AutoModelForVision2Seq, AutoModelForSeq2SeqLM, AutoModelForCausalLM, AutoModel, \
     PretrainedConfig, TextStreamer, LogitsProcessor
+from huggingface_hub import hf_api
+from huggingface_hub.file_download import hf_hub_download
 from transformers.models.auto.modeling_auto import MODEL_FOR_VISION_2_SEQ_MAPPING_NAMES, \
     MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES, MODEL_FOR_CAUSAL_LM_MAPPING_NAMES
 
@@ -25,7 +28,7 @@ from .. import model_management
 from ..component_model.tensor_types import RGBImageBatch
 from ..model_downloader import get_or_download_huggingface_repo
 from ..model_management import unet_offload_device, get_torch_device, unet_dtype, load_models_gpu
-from ..model_management_types import ModelManageable
+from ..model_management_types import ModelManageableStub
 from ..utils import comfy_tqdm, ProgressBar, comfy_progress, seed_for_block
 
 logger = logging.getLogger(__name__)
@@ -37,7 +40,7 @@ _OVERRIDDEN_MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = list(MODEL_FOR_CAUSAL_LM_MAPPING
 _DO_NOT_SKIP_SPECIAL_TOKENS = {'florence2', 'paligemma'}
 
 
-class TransformersManagedModel(ModelManageable, LanguageModel):
+class TransformersManagedModel(ModelManageableStub, LanguageModel):
     def __init__(
             self,
             repo_id: str,
@@ -69,7 +72,20 @@ class TransformersManagedModel(ModelManageable, LanguageModel):
             hub_kwargs["subfolder"] = subfolder
         repo_id = ckpt_name
         with comfy_tqdm():
-            ckpt_name = get_or_download_huggingface_repo(ckpt_name)
+            ckpt_name = get_or_download_huggingface_repo(repo_id)
+
+            if config_dict is None:
+                config_dict, _ = PretrainedConfig.get_config_dict(ckpt_name, **hub_kwargs)
+            elif isinstance(config_dict, PretrainedConfig):
+                config_dict: dict = config_dict.to_dict()
+            else:
+                config_dict = {}
+
+            try:
+                model_type = config_dict["model_type"]
+            except KeyError:
+                logger.debug(f"Configuration was missing for repo_id={repo_id}")
+                model_type = ""
 
             from_pretrained_kwargs = {
                 "pretrained_model_name_or_path": ckpt_name,
@@ -77,19 +93,8 @@ class TransformersManagedModel(ModelManageable, LanguageModel):
                 **hub_kwargs
             }
 
-            # compute bitsandbytes configuration
-            try:
-                import bitsandbytes
-            except ImportError:
-                pass
-
-            if config_dict is None:
-                config_dict, _ = PretrainedConfig.get_config_dict(ckpt_name, **hub_kwargs)
-            elif isinstance(config_dict, PretrainedConfig):
-                config_dict: dict = config_dict.to_dict()
-            model_type = config_dict["model_type"]
             # language models prefer to use bfloat16 over float16
-            kwargs_to_try = ({"torch_dtype": unet_dtype(supported_dtypes=(torch.bfloat16, torch.float16, torch.float32)),
+            kwargs_to_try = ({"dtype": unet_dtype(supported_dtypes=(torch.bfloat16, torch.float16, torch.float32)),
                               "low_cpu_mem_usage": True,
                               "device_map": str(unet_offload_device()), }, {})
 

comfy/lora_types.py

@@ -1,5 +1,5 @@
 from __future__ import annotations
-from typing import Protocol, List, Dict, Optional, NamedTuple, Callable, Literal, Any, TypeAlias, Union
+from typing import Protocol, List, Dict, Optional, NamedTuple, Callable, Literal, Any, TypeAlias, Union, runtime_checkable
 import torch
 
 PatchOffset = tuple[int, int, int]
@@ -31,6 +31,7 @@ class PatchTuple(NamedTuple):
 ModelPatchesDictValue: TypeAlias = list[Union[PatchTuple, PatchWeightTuple]]
 
 
+@runtime_checkable
 class PatchSupport(Protocol):
     """
     Defines the interface for a model that supports LoRA patching.

comfy/model_downloader.py

@@ -45,7 +45,7 @@ def get_filename_list_with_downloadable(folder_name: str, known_files: Optional[
         known_files = _get_known_models_for_folder_name(folder_name)
 
     existing = folder_paths.get_filename_list(folder_name)
-    
+
     downloadable_files = []
     if not args.disable_known_models:
         downloadable_files = known_files
@@ -725,15 +725,15 @@ def get_huggingface_repo_list(*extra_cache_dirs: str) -> List[str]:
         return list(existing_repo_ids | existing_local_dir_repos | known_repo_ids)
 
 
-def get_or_download_huggingface_repo(repo_id: str, cache_dirs: Optional[list] = None, local_dirs: Optional[list] = None) -> Optional[str]:
+def get_or_download_huggingface_repo(repo_id: str, cache_dirs: Optional[list] = None, local_dirs: Optional[list] = None, force: bool = False, subset: bool = False) -> Optional[str]:
     with comfy_tqdm():
-        return _get_or_download_huggingface_repo(repo_id, cache_dirs, local_dirs)
+        return _get_or_download_huggingface_repo(repo_id, cache_dirs, local_dirs, force=force, subset=subset)
 
 
-def _get_or_download_huggingface_repo(repo_id: str, cache_dirs: Optional[list] = None, local_dirs: Optional[list] = None) -> Optional[str]:
+def _get_or_download_huggingface_repo(repo_id: str, cache_dirs: Optional[list] = None, local_dirs: Optional[list] = None, force: bool = False, subset: bool = False) -> Optional[str]:
     cache_dirs = cache_dirs or folder_paths.get_folder_paths("huggingface_cache")
     local_dirs = local_dirs or folder_paths.get_folder_paths("huggingface")
-    cache_dirs_snapshots, local_dirs_snapshots = _get_cache_hits(cache_dirs, local_dirs, repo_id)
+    cache_dirs_snapshots, local_dirs_snapshots = _get_cache_hits(cache_dirs, local_dirs, repo_id, subset=subset)
 
     local_dirs_cache_hit = len(local_dirs_snapshots) > 0
     cache_dirs_cache_hit = len(cache_dirs_snapshots) > 0
@@ -742,25 +742,25 @@ def _get_or_download_huggingface_repo(repo_id: str, cache_dirs: Optional[list] =
     # if we're in forced local directory mode, only use the local dir snapshots, and otherwise, download
     if args.force_hf_local_dir_mode:
         # todo: we still have to figure out a way to download things to the right places by default
-        if len(local_dirs_snapshots) > 0:
+        if len(local_dirs_snapshots) > 0 and not force:
             return local_dirs_snapshots[0]
         elif not args.disable_known_models:
             destination = os.path.join(local_dirs[0], repo_id)
             logger.debug(f"downloading repo_id={repo_id}, local_dir={destination}")
-            return snapshot_download(repo_id, local_dir=destination)
+            return snapshot_download(repo_id, local_dir=destination, force_download=force)
 
     snapshots = local_dirs_snapshots + cache_dirs_snapshots
-    if len(snapshots) > 0:
+    if len(snapshots) > 0 and not force:
         return snapshots[0]
     elif not args.disable_known_models:
         logger.debug(f"downloading repo_id={repo_id}")
-        return snapshot_download(repo_id)
+        return snapshot_download(repo_id, force_download=force)
 
     # this repo was not found
     return None
 
 
-def _get_cache_hits(cache_dirs: Sequence[str], local_dirs: Sequence[str], repo_id):
+def _get_cache_hits(cache_dirs: Sequence[str], local_dirs: Sequence[str], repo_id, subset=False):
     local_dirs_snapshots = []
     cache_dirs_snapshots = []
     # find all the pre-existing downloads for this repo_id
@@ -772,13 +772,12 @@ def _get_cache_hits(cache_dirs: Sequence[str], local_dirs: Sequence[str], repo_i
     if len(repo_files) > 0:
         for local_dir in local_dirs:
             local_path = Path(local_dir) / repo_id
-            local_files = set(f"{repo_id}/{f.relative_to(local_path)}" for f in local_path.rglob("*") if f.is_file())
+            local_files = frozenset(f"{repo_id}/{f.relative_to(local_path)}" for f in local_path.rglob("*") if f.is_file())
             # fix path representation
-            local_files = set(f.replace("\\", "/") for f in local_files)
+            local_files = frozenset(f.replace("\\", "/") for f in local_files)
             # remove .huggingface
-            local_files = set(f for f in local_files if not f.startswith(f"{repo_id}/.huggingface") and not f.startswith(f"{repo_id}/.cache"))
+            local_files = frozenset(f for f in local_files if not f.startswith(f"{repo_id}/.huggingface") and not f.startswith(f"{repo_id}/.cache"))
-            # local_files.issubsetof(repo_files)
+            if len(local_files) > 0 and ((subset and local_files.issubset(repo_files)) or (not subset and repo_files.issubset(local_files))):
-            if len(local_files) > 0 and local_files.issubset(repo_files):
                 local_dirs_snapshots.append(str(local_path))
     else:
         # an empty repository or unknown repository info, trust that if the directory exists, it matches

comfy/model_management_types.py

@@ -1,8 +1,9 @@
 from __future__ import annotations
 
+import copy
 import dataclasses
 from abc import ABCMeta, abstractmethod
-from typing import Protocol, Optional, TypeVar, runtime_checkable, Callable, Any, NamedTuple, override
+from typing import Protocol, Optional, TypeVar, runtime_checkable, Callable, Any, NamedTuple, override, TYPE_CHECKING
 
 import torch
 import torch.nn
@@ -11,22 +12,48 @@ from typing_extensions import TypedDict, NotRequired
 from .comfy_types import UnetWrapperFunction
 from .latent_formats import LatentFormat
 
+if TYPE_CHECKING:
+    from .hooks import EnumHookMode
+
 ModelManageableT = TypeVar('ModelManageableT', bound='ModelManageable')
 LatentFormatT = TypeVar('LatentFormatT', bound=LatentFormat)
 
 
 @runtime_checkable
 class DeviceSettable(Protocol):
-    @property
+    device: torch.device
-    def device(self) -> torch.device:
-        ...
-
-    @device.setter
-    def device(self, value: torch.device):
-        ...
 
 
-class HooksSupport(Protocol, metaclass=ABCMeta):
+@runtime_checkable
+class HooksSupport(Protocol):
+    wrappers: dict[str, dict[str, list[Callable]]]
+    callbacks: dict[str, dict[str, list[Callable]]]
+    hook_mode: "EnumHookMode"
+
+    def prepare_hook_patches_current_keyframe(self, t, hook_group, model_options): ...
+
+    def model_patches_models(self) -> list[ModelManageableT]: ...
+
+    def restore_hook_patches(self): ...
+
+    def cleanup(self): ...
+
+    def pre_run(self): ...
+
+    def prepare_state(self, *args, **kwargs): ...
+
+    def register_all_hook_patches(self, a, b, c, d): ...
+
+    def get_nested_additional_models(self): ...
+
+    def apply_hooks(self, *args, **kwargs): ...
+
+    def add_wrapper(self, wrapper_type: str, wrapper: Callable): ...
+
+    def add_wrapper_with_key(self, wrapper_type: str, key: str, wrapper: Callable): ...
+
+
+class HooksSupportStub(HooksSupport, metaclass=ABCMeta):
     def prepare_hook_patches_current_keyframe(self, t, hook_group, model_options):
         return
 
@@ -82,6 +109,8 @@ class HooksSupport(Protocol, metaclass=ABCMeta):
             if isinstance(model, BaseModel) or hasattr(model, "current_patcher") and isinstance(self, ModelManageable):
                 model.current_patcher = self
 
+
+
     def prepare_state(self, *args, **kwargs):
         pass
 
@@ -94,8 +123,22 @@ class HooksSupport(Protocol, metaclass=ABCMeta):
     def apply_hooks(self, *args, **kwargs):
         return {}
 
+    def add_wrapper(self, wrapper_type: str, wrapper: Callable):
+        self.add_wrapper_with_key(wrapper_type, None, wrapper)
 
-class TrainingSupport(Protocol, metaclass=ABCMeta):
+    def add_wrapper_with_key(self, wrapper_type: str, key: str, wrapper: Callable):
+        w = self.wrappers.setdefault(wrapper_type, {}).setdefault(key, [])
+        w.append(wrapper)
+
+
+@runtime_checkable
+class TrainingSupport(Protocol):
+    def set_model_compute_dtype(self, dtype: torch.dtype): ...
+
+    def add_weight_wrapper(self, name, function): ...
+
+
+class TrainingSupportStub(TrainingSupport, metaclass=ABCMeta):
     def set_model_compute_dtype(self, dtype: torch.dtype):
         return
 
@@ -103,13 +146,68 @@ class TrainingSupport(Protocol, metaclass=ABCMeta):
         return
 
 
-class ModelManageableExtras(Protocol, metaclass=ABCMeta):
+@runtime_checkable
+class ModelManageable(HooksSupport, TrainingSupport, Protocol):
+    """
+    Objects which implement this protocol can be managed by
+
+    >>> from comfy.model_management import load_models_gpu
+    >>> class ModelWrapper(ModelManageable):
+    >>>     ...
+    >>>
+    >>> some_model = ModelWrapper()
+    >>> load_models_gpu([some_model])
+
+    The minimum required
+    """
+    load_device: torch.device
+    offload_device: torch.device
+    model: torch.nn.Module
+
     @property
-    def current_device(self) -> torch.device:
+    def current_device(self) -> torch.device: ...
-        return torch.device("cpu")
+
+    def is_clone(self, other: ModelManageableT) -> bool: ...
+
+    def clone_has_same_weights(self, clone: ModelManageableT) -> bool: ...
+
+    def model_size(self) -> int: ...
+
+    def model_patches_to(self, arg: torch.device | torch.dtype): ...
+
+    def model_dtype(self) -> torch.dtype: ...
+
+    def lowvram_patch_counter(self) -> int: ...
+
+    def partially_load(self, device_to: torch.device, extra_memory: int = 0, force_patch_weights: bool = False) -> int:  ...
+
+    def partially_unload(self, device_to: torch.device, memory_to_free: int = 0) -> int: ...
+
+    def memory_required(self, input_shape: torch.Size) -> int: ...
+
+    def loaded_size(self) -> int: ...
+
+    def current_loaded_device(self) -> torch.device: ...
+
+    def get_model_object(self, name: str) -> torch.nn.Module: ...
+
+    @property
+    def model_options(self) -> ModelOptions: ...
+
+    @model_options.setter
+    def model_options(self, value): ...
+
+    def __del__(self): ...
+
+    @property
+    def parent(self) -> ModelManageableT | None: ...
+
+    def detach(self, unpatch_all: bool = True): ...
+
+    def clone(self) -> ModelManageableT: ...
 
 
-class ModelManageableRequired(Protocol, metaclass=ABCMeta):
+class ModelManageableStub(HooksSupportStub, TrainingSupportStub, ModelManageable, metaclass=ABCMeta):
     """
     The bare minimum that must be implemented to support model management when inheriting from ModelManageable
 
@@ -120,12 +218,11 @@ class ModelManageableRequired(Protocol, metaclass=ABCMeta):
     :see: ModelManageable
     :see: PatchSupport
     """
-    load_device: torch.device
+
-    offload_device: torch.device
-    model: torch.nn.Module
 
     @abstractmethod
-    def patch_model(self, device_to: torch.device | None = None, lowvram_model_memory: int = 0, load_weights: bool = True, force_patch_weights: bool = False) -> torch.nn.Module:
+    def patch_model(self, device_to: torch.device | None = None, lowvram_model_memory: int = 0, load_weights: bool = True,
+                    force_patch_weights: bool = False) -> torch.nn.Module:
         """
         Called by ModelManageable
 
@@ -155,25 +252,6 @@ class ModelManageableRequired(Protocol, metaclass=ABCMeta):
         """
         ...
 
-
-@runtime_checkable
-class ModelManageable(ModelManageableRequired, ModelManageableExtras, Protocol, metaclass=ABCMeta):
-    """
-    Objects which implement this protocol can be managed by
-
-    >>> from comfy.model_management import load_models_gpu
-    >>> class ModelWrapper(ModelManageable):
-    >>>     ...
-    >>>
-    >>> some_model = ModelWrapper()
-    >>> load_models_gpu([some_model])
-
-    The minimum required
-    """
-    load_device: torch.device
-    offload_device: torch.device
-    model: torch.nn.Module
-
     @property
     @override
     def current_device(self) -> torch.device:
@@ -265,6 +343,9 @@ class ModelManageable(ModelManageableRequired, ModelManageableExtras, Protocol,
         self.unpatch_model(self.offload_device, unpatch_weights=unpatch_all)
         return self.model
 
+    def clone(self) -> ModelManageableT:
+        return copy.copy(self)
+
 
 @dataclasses.dataclass
 class MemoryMeasurements:

comfy/model_patcher.py

@@ -230,7 +230,7 @@ class GGUFQuantization:
     patch_on_device: bool = False
 
 
-class ModelPatcher(ModelManageable, TrainingSupport, HooksSupport, PatchSupport):
+class ModelPatcher(ModelManageable, PatchSupport):
     def __init__(self, model: BaseModel | torch.nn.Module, load_device: torch.device, offload_device: torch.device, size=0, weight_inplace_update=False, ckpt_name: Optional[str] = None):
         self.size = size
         self.model: BaseModel | torch.nn.Module = model

comfy/utils.py

@@ -27,6 +27,7 @@ import os
 import random
 import struct
 import sys
+import threading
 import warnings
 from contextlib import contextmanager
 from pathlib import Path
@@ -57,6 +58,9 @@ DISABLE_MMAP = args.disable_mmap
 logger = logging.getLogger(__name__)
 
 ALWAYS_SAFE_LOAD = False
+
+_lock = threading.RLock()
+
 if hasattr(torch.serialization, "add_safe_globals"):  # TODO: this was added in pytorch 2.4, the unsafe path should be removed once earlier versions are deprecated
     class ModelCheckpoint:
         pass
@@ -1178,7 +1182,9 @@ class ProgressBar:
     def update_absolute(self, value, total=None, preview_image_or_output=None):
         if total is not None:
             self.total = total
-        if value > self.total:
+        if value is None:
+            return
+        if self.total is not None and value > self.total:
             value = self.total
         self.current = value
         _progress_bar_update(self.current, self.total, preview_image_or_output, server=self.server, node_id=self.node_id)
@@ -1198,31 +1204,39 @@ def comfy_tqdm() -> Generator[TqdmWatcher, None, None]:
     Monkey patches child calls to tqdm, sends progress to the UI,
     and yields a watcher object for stall detection.
     """
+    with _lock:
+        if hasattr(tqdm, "__patched_by_comfyui__"):
+            yield getattr(tqdm, "__patched_by_comfyui__")
+            return
+
+        watcher = TqdmWatcher()
+        setattr(tqdm, "__patched_by_comfyui__", watcher)
+
     _original_init = tqdm.__init__
     _original_call = tqdm.__call__
     _original_update = tqdm.update
 
     # Create the watcher instance that the patched methods will update
     # and that will be yielded to the caller.
-    watcher = TqdmWatcher()
     context = contextvars.copy_context()
 
     try:
+
         # These inner functions are closures; they capture the `watcher` variable
         # from the enclosing scope.
         def __init(self, *args, **kwargs):
-            context.run(lambda: _original_init(self, *args, **kwargs))
+            _original_init(self, *args, **kwargs)
             self._progress_bar = context.run(lambda: ProgressBar(self.total))
             watcher.tick()  # Signal progress on initialization
 
         def __update(self, n=1):
             assert self._progress_bar is not None
-            context.run(lambda: _original_update(self, n))
+            _original_update(self, n)
             context.run(lambda: self._progress_bar.update(n))
             watcher.tick()  # Signal progress on update
 
         def __call(self, *args, **kwargs):
-            instance = context.run(lambda: _original_call(self, *args, **kwargs))
+            instance = _original_call(self, *args, **kwargs)
             return instance
 
         tqdm.__init__ = __init
@@ -1236,10 +1250,11 @@ def comfy_tqdm() -> Generator[TqdmWatcher, None, None]:
         tqdm.__init__ = _original_init
         tqdm.__call__ = _original_call
         tqdm.update = _original_update
+        delattr(tqdm, "__patched_by_comfyui__")
 
 
 @contextmanager
-def comfy_progress(total: float) -> ProgressBar:
+def comfy_progress(total: float) -> Generator[ProgressBar, Any, None]:
     ctx = current_execution_context()
     if ctx.server.receive_all_progress_notifications:
         yield ProgressBar(total)

comfy_extras/nodes/nodes_custom_sampler.py

@@ -765,7 +765,7 @@ class DualCFGGuider:
     FUNCTION = "get_guider"
     CATEGORY = "sampling/custom_sampling/guiders"
 
-    def get_guider(self, model, cond1, cond2, negative, cfg_conds, cfg_cond2_negative, style):
+    def get_guider(self, model, cond1, cond2, negative, cfg_conds, cfg_cond2_negative, style="regular"):
         guider = Guider_DualCFG(model)
         guider.set_conds(cond1, cond2, negative)
         guider.set_cfg(cfg_conds, cfg_cond2_negative, nested=(style == "nested"))

comfy_extras/nodes/nodes_group_offloading.py

@@ -2,7 +2,9 @@ import torch
 from diffusers import HookRegistry
 from diffusers.hooks import apply_group_offloading, apply_layerwise_casting, ModelHook
 
+from comfy.language.transformers_model_management import TransformersManagedModel
 from comfy.model_management import vram_state, VRAMState
+from comfy.model_management_types import HooksSupport, ModelManageable
 from comfy.model_patcher import ModelPatcher
 from comfy.node_helpers import export_custom_nodes
 from comfy.nodes.package_typing import CustomNode
@@ -117,9 +119,21 @@ class GroupOffload(CustomNode):
     RETURN_TYPES = ("MODEL",)
     FUNCTION = "execute"
 
-    def execute(self, model: ModelPatcher) -> tuple[ModelPatcher,]:
+    def execute(self, model: ModelManageable | HooksSupport | TransformersManagedModel) -> tuple[ModelPatcher,]:
-        model = model.clone()
+        if isinstance(model, ModelManageable):
-        model.add_wrapper(WrappersMP.PREPARE_SAMPLING, prepare_group_offloading_factory(model.load_device, model.offload_device))
+            model = model.clone()
+        if isinstance(model, TransformersManagedModel):
+            apply_group_offloading(
+                model.model,
+                model.load_device,
+                model.offload_device,
+                use_stream=True,
+                record_stream=True,
+                low_cpu_mem_usage=vram_state in (VRAMState.LOW_VRAM,),
+                num_blocks_per_group=1
+            )
+        elif isinstance(model, HooksSupport) and isinstance(model, ModelManageable):
+            model.add_wrapper(WrappersMP.PREPARE_SAMPLING, prepare_group_offloading_factory(model.load_device, model.offload_device))
         return model,
 
 

comfy_extras/nodes/nodes_lt.py

@@ -42,6 +42,7 @@ class LTXVImgToVideo:
                              "height": ("INT", {"default": 512, "min": 64, "max": nodes.MAX_RESOLUTION, "step": 32}),
                              "length": ("INT", {"default": 97, "min": 9, "max": nodes.MAX_RESOLUTION, "step": 8}),
                              "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}),
+                             }, "optional": {
                              "strength": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0}),
                              }}
 
@@ -51,7 +52,7 @@ class LTXVImgToVideo:
     CATEGORY = "conditioning/video_models"
     FUNCTION = "generate"
 
-    def generate(self, positive, negative, image, vae, width, height, length, batch_size, strength):
+    def generate(self, positive, negative, image, vae, width, height, length, batch_size, strength=1.0):
         pixels = comfy.utils.common_upscale(image.movedim(-1, 1), width, height, "bilinear", "center").movedim(1, -1)
         encode_pixels = pixels[:, :, :, :3]
         t = vae.encode(encode_pixels)

comfy_extras/nodes/nodes_upscale_model.py

@@ -9,7 +9,7 @@ from comfy import utils
 from comfy.component_model.tensor_types import RGBImageBatch
 from comfy.model_downloader import get_filename_list_with_downloadable, KNOWN_UPSCALERS, get_or_download
 from comfy.model_management import load_models_gpu
-from comfy.model_management_types import ModelManageable
+from comfy.model_management_types import ModelManageableStub
 
 logger = logging.getLogger(__name__)
 try:
@@ -22,7 +22,7 @@ except:
     pass
 
 
-class UpscaleModelManageable(ModelManageable):
+class UpscaleModelManageable(ModelManageableStub):
     def __init__(self, model_descriptor: ImageModelDescriptor, ckpt_name: str):
         self.ckpt_name = ckpt_name
         self.model_descriptor = model_descriptor

tests/downloader/test_huggingface_downloads.py

@@ -182,7 +182,7 @@ async def test_huggingface_alternate_filenames_in_combo():
     )
 
     # 3. Get the list of files as the UI would
-    filename_list = get_filename_list_with_downloadable("checkpoints", known_files=[known_file])
+    filename_list = get_filename_list_with_downloadable("__xxx___", known_files=[known_file])
 
     # 4. Assert that both the main and alternate filenames are present
     assert main_filename in filename_list

tests/inference/workflows/phi-4-0.json

@@ -1,7 +1,7 @@
 {
   "1": {
     "inputs": {
-      "ckpt_name": "microsoft/Phi-3-mini-4k-instruct",
+      "ckpt_name": "microsoft/Phi-4-mini-instruct",
       "subfolder": ""
     },
     "class_type": "TransformersLoader",
@@ -33,7 +33,7 @@
   "4": {
     "inputs": {
       "prompt": "What comes after apple?",
-      "chat_template": "phi-3",
+      "chat_template": "default",
       "model": [
         "1",
         0