Improvements to GGUF

- GGUF now works and included 4 bit quants, this will allow WAN to run on 24GB VRAM GPUs
 - logger only shows full stack for errors
 - helper functions for colab notebook
 - fix nvcr.io auth error
 - lora issues are now reported better
 - model downloader will use huggingface cache and symlinking if it is supported on your platform
 - torch compile node now correctly patches the model before compilation

.github/workflows/docker-build.yml

@@ -25,6 +25,11 @@ jobs:
           registry: ${{ env.REGISTRY }}
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
+      - uses: docker/login-action@v3
+        with:
+          registry: nvcr.io
+          username: "$oauthtoken"
+          password: ${{ secrets.NVCR_NGC_TOKEN }}
       - name: Build and push CUDA (NVIDIA) image
         uses: docker/build-push-action@v6
         with:

comfy/app/colab.py

@@ -0,0 +1,138 @@
+import asyncio
+import os
+import stat
+import subprocess
+import threading
+from asyncio import Task
+from typing import NamedTuple
+
+import requests
+
+from ..cmd.folder_paths import init_default_paths, folder_names_and_paths
+# experimental workarounds for colab
+from ..cmd.main import _start_comfyui
+from ..execution_context import *
+
+
+class _ColabTuple(NamedTuple):
+    tunnel: "CloudflaredTunnel"
+    server: Task
+
+
+_colab_instances: list[_ColabTuple] = []
+
+
+class CloudflaredTunnel:
+    """
+    A class to manage a cloudflared tunnel subprocess.
+
+    Provides methods to start, stop, and manage the lifecycle of the tunnel.
+    It can be used as a context manager.
+    """
+
+    def __init__(self, port: int):
+        self._port: int = port
+        self._executable_path: str = "./cloudflared"
+        self._process: Optional[subprocess.Popen] = None
+        self._thread: Optional[threading.Thread] = None
+
+        # Download and set permissions for the executable
+        self._setup_executable()
+
+        # Start the tunnel process and capture the URL
+        self.url: str = self._start_tunnel()
+
+    def _setup_executable(self):
+        """Downloads cloudflared and makes it executable if it doesn't exist."""
+        if not os.path.exists(self._executable_path):
+            url = "https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64"
+            response = requests.get(url, stream=True)
+            response.raise_for_status()
+            with open(self._executable_path, "wb") as f:
+                for chunk in response.iter_content(chunk_size=8192):
+                    f.write(chunk)
+
+            # Make the file executable (add execute permission for the owner)
+            current_permissions = os.stat(self._executable_path).st_mode
+            os.chmod(self._executable_path, current_permissions | stat.S_IEXEC)
+
+    def _start_tunnel(self) -> str:
+        """Starts the tunnel and returns the public URL."""
+        command = [self._executable_path, "tunnel", "--url", f"http://localhost:{self._port}", "--no-autoupdate"]
+
+        # Using DEVNULL for stderr to keep the output clean, stdout is piped
+        self._process = subprocess.Popen(
+            command,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.DEVNULL,
+            text=True
+        )
+
+        for line in iter(self._process.stdout.readline, ""):
+            if ".trycloudflare.com" in line:
+                # The line format is typically: "INFO |  https://<subdomain>.trycloudflare.com |"
+                try:
+                    url = line.split("|")[1].strip()
+                    print(f"Tunnel is live at: {url}")
+                    return url
+                except IndexError:
+                    continue
+
+        # If the loop finishes without finding a URL
+        self.stop()
+        raise RuntimeError("Failed to start cloudflared tunnel or find URL.")
+
+    def stop(self):
+        """Stops the cloudflared tunnel process."""
+        if self._process and self._process.poll() is None:
+            print("Stopping cloudflared tunnel...")
+            self._process.terminate()
+            try:
+                self._process.wait(timeout=5)
+                print("Tunnel stopped successfully.")
+            except subprocess.TimeoutExpired:
+                print("Tunnel did not terminate gracefully, forcing kill.")
+                self._process.kill()
+        self._process = None
+
+    def __enter__(self):
+        """Enter context manager."""
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """Exit context manager, ensuring the tunnel is stopped."""
+        self.stop()
+
+
+def start_tunnel(port: int) -> CloudflaredTunnel:
+    """
+    Initializes and starts a cloudflared tunnel.
+
+    Args:
+        port: The local port number to expose.
+
+    Returns:
+        A CloudflaredTunnel object that controls the tunnel process.
+        This object has a `url` attribute and a `stop()` method.
+    """
+    return CloudflaredTunnel(port)
+
+
+def start_server_in_colab() -> str:
+    """
+    returns the URL of the tunnel and the running context
+    :return:
+    """
+    if len(_colab_instances) == 0:
+        comfyui_execution_context.set(ExecutionContext(server=ServerStub(), folder_names_and_paths=FolderNames(is_root=True), custom_nodes=ExportedNodes(), progress_registry=ProgressRegistryStub()))
+
+        async def colab_server_loop():
+            init_default_paths(folder_names_and_paths)
+            await _start_comfyui()
+
+        _loop = asyncio.get_running_loop()
+        task = _loop.create_task(colab_server_loop())
+
+        tunnel = start_tunnel(8188)
+        _colab_instances.append(_ColabTuple(tunnel, task))
+    return _colab_instances[0].tunnel.url

comfy/app/logger.py

@@ -55,7 +55,7 @@ def on_flush(callback):
 
 class StackTraceLogger(logging.Logger):
     def _log(self, level, msg, args, exc_info=None, extra=None, stack_info=False, stacklevel=1):
-        if level >= logging.WARNING:
+        if level >= logging.ERROR:
             stack_info = True
         super()._log(level, msg, args, exc_info, extra, stack_info, stacklevel=stacklevel + 1)
 

comfy/client/embedded_comfy_client.py

@@ -171,6 +171,16 @@ class Comfy:
     def task_count(self) -> int:
         return self._task_count
 
+    def __enter__(self):
+        self._is_running = True
+        return self
+
+    def __exit__(self, *args):
+        get_event_loop().run_in_executor(self._executor, _cleanup)
+        self._executor.shutdown(wait=True)
+        self._is_running = False
+
+
     async def __aenter__(self):
         self._is_running = True
         return self

comfy/cmd/execution.py

@@ -19,15 +19,14 @@ from typing import List, Optional, Tuple, Literal
 import torch
 from opentelemetry.trace import get_current_span, StatusCode, Status
 
-# order matters
-from .main_pre import tracer
-
 from comfy_execution.caching import HierarchicalCache, LRUCache, CacheKeySetInputSignature, CacheKeySetID, \
     DependencyAwareCache, \
     BasicCache
 from comfy_execution.graph import get_input_info, ExecutionList, DynamicPrompt, ExecutionBlocker
 from comfy_execution.graph_utils import is_link, GraphBuilder
 from comfy_execution.utils import CurrentNodeContext
+# order matters
+from .main_pre import tracer
 from .. import interruption
 from .. import model_management
 from ..cli_args import args
@@ -38,7 +37,8 @@ from ..component_model.executor_types import ExecutorToClientProgress, Validatio
     HistoryResultDict, ExecutionErrorMessage, ExecutionInterruptedMessage
 from ..component_model.files import canonicalize_path
 from ..component_model.module_property import create_module_properties
-from ..component_model.queue_types import QueueTuple, HistoryEntry, QueueItem, MAXIMUM_HISTORY_SIZE, ExecutionStatus, ExecutionStatusAsDict
+from ..component_model.queue_types import QueueTuple, HistoryEntry, QueueItem, MAXIMUM_HISTORY_SIZE, ExecutionStatus, \
+    ExecutionStatusAsDict
 from ..execution_context import context_execute_node, context_execute_prompt
 from ..execution_ext import should_panic_on_exception
 from ..nodes.package_typing import InputTypeSpec, FloatSpecOptions, IntSpecOptions, CustomNode
@@ -48,6 +48,7 @@ from ..progress import get_progress_state, reset_progress_state, add_progress_ha
 from ..validation import validate_node_input
 
 _module_properties = create_module_properties()
+logger = logging.getLogger(__name__)
 
 
 @_module_properties.getter
@@ -100,12 +101,12 @@ class CacheSet:
     def __init__(self, cache_type=None, cache_size=None):
         if cache_type == CacheType.DEPENDENCY_AWARE:
             self.init_dependency_aware_cache()
-            logging.info("Disabling intermediate node cache.")
+            logger.info("Disabling intermediate node cache.")
         elif cache_type == CacheType.LRU:
             if cache_size is None:
                 cache_size = 0
             self.init_lru_cache(cache_size)
-            logging.info("Using LRU cache")
+            logger.info("Using LRU cache")
         else:
             self.init_classic_cache()
 
@@ -571,7 +572,7 @@ async def _execute(server, dynprompt, caches: CacheSet, current_item: str, extra
             return RecursiveExecutionTuple(ExecutionResult.PENDING, None, None)
         caches.outputs.set(unique_id, output_data)
     except interruption.InterruptProcessingException as iex:
-        logging.info("Processing interrupted")
+        logger.info("Processing interrupted")
 
         # skip formatting inputs/outputs
         error_details: RecursiveExecutionErrorDetailsInterrupted = {
@@ -588,13 +589,13 @@ async def _execute(server, dynprompt, caches: CacheSet, current_item: str, extra
             for name, inputs in input_data_all.items():
                 input_data_formatted[name] = [format_value(x) for x in inputs]
 
-        logging.error("An error occurred while executing a workflow", exc_info=ex)
-        logging.error(traceback.format_exc())
+        logger.error("An error occurred while executing a workflow", exc_info=ex)
+        logger.error(traceback.format_exc())
         tips = ""
 
         if isinstance(ex, model_management.OOM_EXCEPTION):
             tips = "This error means you ran out of memory on your GPU.\n\nTIPS: If the workflow worked before you might have accidentally set the batch_size to a large number."
-            logging.error("Got an OOM, unloading all loaded models.")
+            logger.error("Got an OOM, unloading all loaded models.")
             model_management.unload_all_models()
 
         error_details: RecursiveExecutionErrorDetails = {
@@ -606,7 +607,7 @@ async def _execute(server, dynprompt, caches: CacheSet, current_item: str, extra
         }
 
         if should_panic_on_exception(ex, args.panic_when):
-            logging.error(f"The exception {ex} was configured as unrecoverable, scheduling an exit")
+            logger.error(f"The exception {ex} was configured as unrecoverable, scheduling an exit")
 
             def sys_exit(*args):
                 sys.exit(1)
@@ -1120,11 +1121,11 @@ async def _validate_prompt(prompt_id: typing.Any, prompt: typing.Mapping[str, ty
         if valid is True:
             good_outputs.add(o)
         else:
-            logging.error(f"Failed to validate prompt for output {o}:")
+            logger.error(f"Failed to validate prompt for output {o}:")
             if len(reasons) > 0:
-                logging.error("* (prompt):")
+                logger.error("* (prompt):")
                 for reason in reasons:
-                    logging.error(f"  - {reason['message']}: {reason['details']}")
+                    logger.error(f"  - {reason['message']}: {reason['details']}")
             errors += [(o, reasons)]
             for node_id, result in validated.items():
                 valid = result[0]
@@ -1140,11 +1141,11 @@ async def _validate_prompt(prompt_id: typing.Any, prompt: typing.Mapping[str, ty
                             "dependent_outputs": [],
                             "class_type": class_type
                         }
-                        logging.error(f"* {class_type} {node_id}:")
+                        logger.error(f"* {class_type} {node_id}:")
                         for reason in reasons:
-                            logging.error(f"  - {reason['message']}: {reason['details']}")
+                            logger.error(f"  - {reason['message']}: {reason['details']}")
                     node_errors[node_id]["dependent_outputs"].append(o)
-            logging.error("Output will be ignored")
+            logger.error("Output will be ignored")
 
     if len(good_outputs) == 0:
         errors_list = []

comfy/cmd/folder_paths.py

@@ -93,7 +93,7 @@ def init_default_paths(folder_names_and_paths: FolderNames, configuration: Optio
         ModelPaths(["checkpoints"], supported_extensions=set(supported_pt_extensions)),
         ModelPaths(["configs"], additional_absolute_directory_paths=[get_package_as_path("comfy.configs")], supported_extensions={".yaml"}),
         ModelPaths(["vae"], supported_extensions=set(supported_pt_extensions)),
-        ModelPaths(folder_names=["clip", "text_encoders"], supported_extensions=set(supported_pt_extensions)),
+        ModelPaths(folder_names=["text_encoders", "clip"], supported_extensions=set(supported_pt_extensions)),
         ModelPaths(["loras"], supported_extensions=set(supported_pt_extensions)),
         ModelPaths(folder_names=["diffusion_models", "unet"], supported_extensions=set(supported_pt_extensions), folder_names_are_relative_directory_paths_too=True),
         ModelPaths(["clip_vision"], supported_extensions=set(supported_pt_extensions)),

comfy/gguf.py

@@ -939,16 +939,16 @@ def get_torch_compiler_disable_decorator():
     from packaging import version
 
     if not chained_hasattr(torch, "compiler.disable"):
-        logger.info("ComfyUI-GGUF: Torch too old for torch.compile - bypassing")
+        logger.debug("ComfyUI-GGUF: Torch too old for torch.compile - bypassing")
         return dummy_decorator  # torch too old
     elif version.parse(torch.__version__) >= version.parse("2.8"):
-        logger.info("ComfyUI-GGUF: Allowing full torch compile")
+        logger.debug("ComfyUI-GGUF: Allowing full torch compile")
         return dummy_decorator  # torch compile works
     if chained_hasattr(torch, "_dynamo.config.nontraceable_tensor_subclasses"):
-        logger.info("ComfyUI-GGUF: Allowing full torch compile (nightly)")
+        logger.debug("ComfyUI-GGUF: Allowing full torch compile (nightly)")
         return dummy_decorator  # torch compile works, nightly before 2.8 release
     else:
-        logger.info("ComfyUI-GGUF: Partial torch compile only, consider updating pytorch")
+        logger.debug("ComfyUI-GGUF: Partial torch compile only, consider updating pytorch")
         return torch.compiler.disable
 
 

comfy/lora.py

@@ -24,8 +24,10 @@ import torch
 from . import model_base
 from . import model_management
 from . import utils
-from .lora_types import PatchDict, PatchOffset, PatchConversionFunction, PatchType, ModelPatchesDictValue
 from . import weight_adapter
+from .lora_types import PatchDict, PatchOffset, PatchConversionFunction, PatchType, ModelPatchesDictValue
+
+logger = logging.getLogger(__name__)
 
 LORA_CLIP_MAP = {
     "mlp.fc1": "mlp_fc1",
@@ -37,7 +39,7 @@ LORA_CLIP_MAP = {
 }
 
 
-def load_lora(lora, to_load, log_missing=True) -> PatchDict:
+def load_lora(lora, to_load, log_missing=True, lora_name=None) -> PatchDict:
     patch_dict: PatchDict = {}
     loaded_keys = set()
     for x in to_load:
@@ -91,9 +93,10 @@ def load_lora(lora, to_load, log_missing=True) -> PatchDict:
             loaded_keys.add(set_weight_name)
 
     if log_missing:
-        for x in lora.keys():
-            if x not in loaded_keys:
-                logging.warning("lora key not loaded: {}".format(x))
+        not_loaded_keys = [x for x in lora.keys() if x not in loaded_keys]
+        n_not_loaded_keys = len(not_loaded_keys)
+        if n_not_loaded_keys > 0:
+            logger.warning(f"[{lora_name}] lora keys not loaded ({n_not_loaded_keys} / {len(loaded_keys) + n_not_loaded_keys}): {not_loaded_keys}")
 
     return patch_dict
 
@@ -293,12 +296,12 @@ def model_lora_keys_unet(model, key_map=None):
             if k.startswith("diffusion_model."):
                 if k.endswith(".weight"):
                     key_lora = k[len("diffusion_model."):-len(".weight")]
-                    key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k #SimpleTuner lycoris format
-                    key_map["transformer.{}".format(key_lora)] = k #SimpleTuner regular format
+                    key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k  # SimpleTuner lycoris format
+                    key_map["transformer.{}".format(key_lora)] = k  # SimpleTuner regular format
 
     if isinstance(model, model_base.ACEStep):
         for k in sdk:
-            if k.startswith("diffusion_model.") and k.endswith(".weight"): #Official ACE step lora format
+            if k.startswith("diffusion_model.") and k.endswith(".weight"):  # Official ACE step lora format
                 key_lora = k[len("diffusion_model."):-len(".weight")]
                 key_map["{}".format(key_lora)] = k
 
@@ -364,7 +367,7 @@ def calculate_weight(patches: ModelPatchesDictValue, weight, key, intermediate_d
         if isinstance(v, weight_adapter.WeightAdapterBase):
             output = v.calculate_weight(weight, key, strength, strength_model, offset, function, intermediate_dtype, original_weights)
             if output is None:
-                logging.warning("Calculate Weight Failed: {} {}".format(v.name, key))
+                logger.warning("Calculate Weight Failed: {} {}".format(v.name, key))
             else:
                 weight = output
                 if old_weight is not None:
@@ -382,12 +385,12 @@ def calculate_weight(patches: ModelPatchesDictValue, weight, key, intermediate_d
             # An extra flag to pad the weight if the diff's shape is larger than the weight
             do_pad_weight = len(v) > 1 and v[1]['pad_weight']
             if do_pad_weight and diff.shape != weight.shape:
-                logging.info("Pad weight {} from {} to shape: {}".format(key, weight.shape, diff.shape))
+                logger.info("Pad weight {} from {} to shape: {}".format(key, weight.shape, diff.shape))
                 weight = pad_tensor_to_shape(weight, diff.shape)
 
             if strength != 0.0:
                 if diff.shape != weight.shape:
-                    logging.warning("WARNING SHAPE MISMATCH {} WEIGHT NOT MERGED {} != {}".format(key, diff.shape, weight.shape))
+                    logger.warning("WARNING SHAPE MISMATCH {} WEIGHT NOT MERGED {} != {}".format(key, diff.shape, weight.shape))
                 else:
                     weight += function(strength * model_management.cast_to_device(diff, weight.device, weight.dtype))
         elif patch_type == "set":
@@ -398,7 +401,7 @@ def calculate_weight(patches: ModelPatchesDictValue, weight, key, intermediate_d
                           model_management.cast_to_device(original_weights[key][0][0], weight.device, intermediate_dtype)
             weight += function(strength * model_management.cast_to_device(diff_weight, weight.device, weight.dtype))
         else:
-            logging.warning("patch type not recognized {} {}".format(patch_type, key))
+            logger.warning("patch type not recognized {} {}".format(patch_type, key))
 
         if old_weight is not None:
             weight = old_weight

comfy/model_downloader.py

@@ -1,8 +1,5 @@
 from __future__ import annotations
 
-from itertools import chain
-from os.path import join
-
 import collections
 import logging
 import operator
@@ -10,6 +7,8 @@ import os
 import shutil
 from collections.abc import Sequence, MutableSequence
 from functools import reduce
+from itertools import chain
+from os.path import join
 from pathlib import Path
 from typing import List, Optional, Final, Set
 
@@ -66,28 +65,38 @@ def get_or_download(folder_name: str, filename: str, known_files: Optional[List[
     filename = canonicalize_path(filename)
     path = folder_paths.get_full_path(folder_name, filename)
 
+    candidate_str_match = False
+    candidate_filename_match = False
+    candidate_alternate_filenames_match = False
+    candidate_save_filename_match = False
     if path is None and not args.disable_known_models:
         try:
             # todo: should this be the first or last path?
             this_model_directory = folder_paths.get_folder_paths(folder_name)[0]
             known_file: Optional[HuggingFile | CivitFile] = None
             for candidate in known_files:
-                if (canonicalize_path(str(candidate)) == filename
-                        or canonicalize_path(candidate.filename) == filename
-                        or filename in list(map(canonicalize_path, candidate.alternate_filenames))
-                        or filename == canonicalize_path(candidate.save_with_filename)):
+                candidate_str_match = canonicalize_path(str(candidate)) == filename
+                candidate_filename_match = canonicalize_path(candidate.filename) == filename
+                candidate_alternate_filenames_match = filename in list(map(canonicalize_path, candidate.alternate_filenames))
+                candidate_save_filename_match = filename == canonicalize_path(candidate.save_with_filename)
+                if (candidate_str_match
+                        or candidate_filename_match
+                        or candidate_alternate_filenames_match
+                        or candidate_save_filename_match):
                     known_file = candidate
                     break
             if known_file is None:
+                logger.debug(f"get_or_download could not find {filename} in {folder_name}, known_files={known_files}")
                 return path
             with comfy_tqdm():
                 if isinstance(known_file, HuggingFile):
+                    symlinks_supported = are_symlinks_supported()
                     if known_file.save_with_filename is not None:
                         linked_filename = known_file.save_with_filename
                     elif not known_file.force_save_in_repo_id and os.path.basename(known_file.filename) != known_file.filename:
                         linked_filename = os.path.basename(known_file.filename)
                     else:
-                        linked_filename = None
+                        linked_filename = known_file.filename
 
                     if known_file.force_save_in_repo_id or linked_filename is not None and os.path.dirname(known_file.filename) == "":
                         # if the known file has an overridden linked name, save it into a repo_id sub directory
@@ -112,8 +121,6 @@ def get_or_download(folder_name: str, filename: str, known_files: Optional[List[
 
                     cache_hit = False
                     try:
-                        if not are_symlinks_supported():
-                            raise PermissionError("no symlink support")
                         # always retrieve this from the cache if it already exists there
                         path = hf_hub_download(repo_id=known_file.repo_id,
                                                filename=known_file.filename,
@@ -121,19 +128,20 @@ def get_or_download(folder_name: str, filename: str, known_files: Optional[List[
                                                revision=known_file.revision,
                                                local_files_only=True,
                                                )
-                        logger.info(f"hf_hub_download cache hit for {known_file.repo_id}/{known_file.filename}")
-                        if linked_filename is None:
-                            linked_filename = known_file.filename
+                        logger.debug(f"hf_hub_download cache hit for {known_file.repo_id}/{known_file.filename}")
                         cache_hit = True
-                    except (LocalEntryNotFoundError, PermissionError):
-                        path = hf_hub_download(repo_id=known_file.repo_id,
-                                               filename=known_file.filename,
-                                               local_dir=hf_destination_dir,
-                                               repo_type=known_file.repo_type,
-                                               revision=known_file.revision,
-                                               )
+                    except LocalEntryNotFoundError:
+                        try:
+                            logger.debug(f"{folder_name}/{filename} is being downloaded from {known_file.repo_id}/{known_file.filename} candidate_str_match={candidate_str_match} candidate_filename_match={candidate_filename_match} candidate_alternate_filenames_match={candidate_alternate_filenames_match} candidate_save_filename_match={candidate_save_filename_match}")
+                            path = hf_hub_download(repo_id=known_file.repo_id,
+                                                   filename=known_file.filename,
+                                                   repo_type=known_file.repo_type,
+                                                   revision=known_file.revision,
+                                                   )
+                        except IOError as exc_info:
+                            logger.error(f"cannot reach huggingface {known_file.repo_id}/{known_file.filename}", exc_info=exc_info)
 
-                    if known_file.convert_to_16_bit and file_size is not None and file_size != 0:
+                    if path is not None and known_file.convert_to_16_bit and file_size is not None and file_size != 0:
                         tensors = {}
                         with safe_open(path, framework="pt") as f:
                             with tqdm.tqdm(total=len(f.keys())) as pb:
@@ -151,20 +159,23 @@ def get_or_download(folder_name: str, filename: str, known_files: Optional[List[
                         logger.info(f"Converted {path} to 16 bit, size is {os.stat(path, follow_symlinks=True).st_size}")
 
                     link_successful = True
-                    if linked_filename is not None:
+                    if path is not None:
                         destination_link = os.path.join(this_model_directory, linked_filename)
-                        try:
-                            os.makedirs(this_model_directory, exist_ok=True)
-                            os.symlink(path, destination_link)
-                        except Exception as exc_info:
-                            logger.error("error while symbolic linking", exc_info=exc_info)
+                        if Path(destination_link).is_file():
+                            logger.warning(f"{known_file.repo_id}/{known_file.filename} could not link to {destination_link} because the path already exists, which is unexpected")
+                        else:
                             try:
-                                os.link(path, destination_link)
-                            except Exception as hard_link_exc:
-                                logger.error("error while hard linking", exc_info=hard_link_exc)
-                                if cache_hit:
-                                    shutil.copyfile(path, destination_link)
-                                link_successful = False
+                                os.makedirs(this_model_directory, exist_ok=True)
+                                os.symlink(path, destination_link)
+                            except Exception as exc_info:
+                                logger.error("error while symbolic linking", exc_info=exc_info)
+                                try:
+                                    os.link(path, destination_link)
+                                except Exception as hard_link_exc:
+                                    logger.error("error while hard linking", exc_info=hard_link_exc)
+                                    if cache_hit:
+                                        shutil.copyfile(path, destination_link)
+                                    link_successful = False
 
                     if not link_successful:
                         logger.error(f"Failed to link file with alternative download save name in a way that is compatible with Hugging Face caching {repr(known_file)}. If cache_hit={cache_hit} is True, the file was copied into the destination.", exc_info=exc_info)
@@ -558,7 +569,9 @@ KNOWN_UNET_MODELS: Final[KnownDownloadables] = KnownDownloadables([
     HuggingFile("Comfy-Org/Wan_2.2_ComfyUI_Repackaged", "split_files/diffusion_models/wan2.2_ti2v_5B_fp16.safetensors"),
     HuggingFile("lodestones/Chroma", "chroma-unlocked-v37.safetensors"),
     HuggingFile("QuantStack/Wan2.2-T2V-A14B-GGUF", "HighNoise/Wan2.2-T2V-A14B-HighNoise-Q8_0.gguf"),
+    HuggingFile("QuantStack/Wan2.2-T2V-A14B-GGUF", "HighNoise/Wan2.2-T2V-A14B-HighNoise-Q4_K_M.gguf"),
     HuggingFile("QuantStack/Wan2.2-T2V-A14B-GGUF", "LowNoise/Wan2.2-T2V-A14B-LowNoise-Q8_0.gguf"),
+    HuggingFile("QuantStack/Wan2.2-T2V-A14B-GGUF", "LowNoise/Wan2.2-T2V-A14B-LowNoise-Q4_K_M.gguf"),
 ], folder_names=["diffusion_models", "unet"])
 
 KNOWN_CLIP_MODELS: Final[KnownDownloadables] = KnownDownloadables([

comfy/model_patcher.py

@@ -630,7 +630,7 @@ class ModelPatcher(ModelManageable):
         # from gguf
         if is_quantized(weight):
             out_weight = weight.to(device_to)
-            patches = move_patch_to_device(self.patches[key], self.load_device if self.patch_on_device else self.offload_device)
+            patches = move_patch_to_device(self.patches[key], self.load_device if self.gguf.patch_on_device else self.offload_device)
             # TODO: do we ever have legitimate duplicate patches? (i.e. patch on top of patched weight)
             out_weight.patches = [(patches, key)]
             if inplace_update:

comfy/nodes/base_nodes.py

@@ -674,7 +674,7 @@ class LoraLoader:
             lora = utils.load_torch_file(lora_path, safe_load=True)
             self.loaded_lora = (lora_path, lora)
 
-        model_lora, clip_lora = sd.load_lora_for_models(model, clip, lora, strength_model, strength_clip)
+        model_lora, clip_lora = sd.load_lora_for_models(model, clip, lora, strength_model, strength_clip, lora_name=lora_name)
         return (model_lora, clip_lora)
 
 class LoraLoaderModelOnly(LoraLoader):

comfy/sd.py

@@ -15,7 +15,6 @@ import yaml
 from . import clip_vision
 from . import diffusers_convert
 from . import gligen
-from . import lora
 from . import model_detection
 from . import model_management
 from . import model_patcher
@@ -37,6 +36,7 @@ from .ldm.lightricks.vae import causal_video_autoencoder as lightricks
 from .ldm.models.autoencoder import AutoencoderKL, AutoencodingEngine
 from .ldm.wan import vae as wan_vae
 from .ldm.wan import vae2_2 as wan_vae2_2
+from .lora import load_lora, model_lora_keys_unet, model_lora_keys_clip
 from .lora_convert import convert_lora
 from .model_management import load_models_gpu
 from .model_patcher import ModelPatcher
@@ -64,15 +64,15 @@ from .utils import ProgressBar, FileMetadata
 logger = logging.getLogger(__name__)
 
 
-def load_lora_for_models(model, clip, _lora, strength_model, strength_clip):
+def load_lora_for_models(model, clip, lora, strength_model, strength_clip, lora_name=None):
     key_map = {}
     if model is not None:
-        key_map = lora.model_lora_keys_unet(model.model, key_map)
+        key_map = model_lora_keys_unet(model.model, key_map)
     if clip is not None:
-        key_map = lora.model_lora_keys_clip(clip.cond_stage_model, key_map)
+        key_map = model_lora_keys_clip(clip.cond_stage_model, key_map)
 
-    _lora = convert_lora(_lora)
-    loaded = lora.load_lora(_lora, key_map)
+    lora = convert_lora(lora)
+    loaded = load_lora(lora, key_map, lora_name=lora_name)
     if model is not None:
         new_modelpatcher: ModelPatcher = model.clone()
         k = new_modelpatcher.add_patches(loaded, strength_model)
@@ -90,7 +90,7 @@ def load_lora_for_models(model, clip, _lora, strength_model, strength_clip):
     k1 = set(k1)
     for x in loaded:
         if (x not in k) and (x not in k1):
-            logger.warning("NOT LOADED {}".format(x))
+            logger.warning(f"[{lora_name}] clip keys not loaded {x}".format(x))
 
     return (new_modelpatcher, new_clip)
 

comfy_extras/nodes/nodes_torch_compile.py

@@ -80,7 +80,6 @@ class TorchCompileModel(CustomNode):
             "backend": backend,
             "mode": mode,
         }
-        move_to_gpu = True
         try:
             if backend == "torch_tensorrt":
                 try:
@@ -98,7 +97,6 @@ class TorchCompileModel(CustomNode):
                     "enable_weight_streaming": True,
                     "make_refittable": True,
                 }
-                move_to_gpu = True
                 del compile_kwargs["mode"]
             if isinstance(model, (ModelPatcher, TransformersManagedModel, VAE)):
                 to_return = model.clone()
@@ -109,27 +107,18 @@ class TorchCompileModel(CustomNode):
                     object_patches = ["encoder", "decoder"]
                 else:
                     patcher = to_return
-                if object_patch is None or len(object_patches) == 0:
+                if object_patch is None or len(object_patches) == 0 or len(object_patches) == 1 and object_patches[0].strip() == "":
                     object_patches = [DIFFUSION_MODEL]
-                if move_to_gpu:
-                    model_management.unload_all_models()
-                    model_management.load_models_gpu([patcher])
                 set_torch_compile_wrapper(patcher, keys=object_patches, **compile_kwargs)
-                # m.add_object_patch(object_patch, torch.compile(model=m.get_model_object(object_patch), **compile_kwargs))
-                # todo: do we want to move something back off the GPU?
-                # if move_to_gpu:
-                #     model_management.unload_all_models()
                 return to_return,
             elif isinstance(model, torch.nn.Module):
-                if move_to_gpu:
-                    model_management.unload_all_models()
-                    model.to(device=model_management.get_torch_device())
+                model_management.unload_all_models()
+                model.to(device=model_management.get_torch_device())
                 res = torch.compile(model=model, **compile_kwargs),
-                if move_to_gpu:
-                    model.to(device=model_management.unet_offload_device())
+                model.to(device=model_management.unet_offload_device())
                 return res,
             else:
-                logger.warning("Encountered a model that cannot be compiled")
+                logger.warning(f"Encountered a model {model} that cannot be compiled")
                 return model,
         except OSError as os_error:
             try:
@@ -174,7 +163,8 @@ class QuantizeModel(CustomNode):
 
     def execute(self, model: ModelPatcher, strategy: str = _QUANTIZATION_STRATEGIES[0]) -> tuple[ModelPatcher]:
         model = model.clone()
-        unet = model.get_model_object("diffusion_model")
+        model.patch_model(force_patch_weights=True)
+        unet = model.diffusion_model
         # todo: quantize quantizes in place, which is not desired
 
         # default exclusions
@@ -209,7 +199,7 @@ class QuantizeModel(CustomNode):
             if "autoquant" in strategy:
                 _in_place_fixme = autoquant(unet, error_on_unseen=False)
             else:
-                quantize_(unet, int8_dynamic_activation_int8_weight(), device=model_management.get_torch_device())
+                quantize_(unet, int8_dynamic_activation_int8_weight(), device=model_management.get_torch_device(), filter_fn=filter)
                 _in_place_fixme = unet
             unwrap_tensor_subclass(_in_place_fixme)
         else:

comfy_extras/nodes/nodes_train.py

@@ -717,7 +717,7 @@ class LoraModelLoader:
         if strength_model == 0:
             return (model, )
 
-        model_lora, _ = comfy.sd.load_lora_for_models(model, None, lora, strength_model, 0)
+        model_lora, _ = comfy.sd.load_lora_for_models(model, None, lora, strength_model, 0, None)
         return (model_lora, )