From fe205a67f49a0892e5caa3c65c7049f43f8609ef Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Tue, 17 Dec 2024 02:15:58 +0000
Subject: [PATCH 01/23] Add no_mmap arg.

---
 src/diffusers/loaders/single_file.py        | 3 +++
 src/diffusers/loaders/single_file_utils.py  | 3 ++-
 src/diffusers/models/model_loading_utils.py | 7 +++++--
 3 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index c0cbfc713857..149be4ebcc04 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -308,6 +308,8 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
                       hosted on the Hub.
                     - A path to a *directory* (for example `./my_pipeline_directory/`) containing the pipeline
                       component configs in Diffusers format.
+            no_mmap ('bool', *optional*, defaults to 'False'):
+                Whether to use mmap when loading the model.
             kwargs (remaining dictionary of keyword arguments, *optional*):
                 Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline
                 class). The overwritten components are passed directly to the pipelines `__init__` method. See example
@@ -383,6 +385,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
             cache_dir=cache_dir,
             local_files_only=local_files_only,
             revision=revision,
+            no_mmap=no_mmap,
         )
 
         if config is None:
diff --git a/src/diffusers/loaders/single_file_utils.py b/src/diffusers/loaders/single_file_utils.py
index 21ff2841700d..24fb3452986d 100644
--- a/src/diffusers/loaders/single_file_utils.py
+++ b/src/diffusers/loaders/single_file_utils.py
@@ -364,6 +364,7 @@ def load_single_file_checkpoint(
     cache_dir=None,
     local_files_only=None,
     revision=None,
+    no_mmap=False,
 ):
     if os.path.isfile(pretrained_model_link_or_path):
         pretrained_model_link_or_path = pretrained_model_link_or_path
@@ -381,7 +382,7 @@ def load_single_file_checkpoint(
             revision=revision,
         )
 
-    checkpoint = load_state_dict(pretrained_model_link_or_path)
+    checkpoint = load_state_dict(pretrained_model_link_or_path, no_mmap)
 
     # some checkpoints contain the model state dict under a "state_dict" key
     while "state_dict" in checkpoint:
diff --git a/src/diffusers/models/model_loading_utils.py b/src/diffusers/models/model_loading_utils.py
index 546c0eb4d840..a4d5c082e433 100644
--- a/src/diffusers/models/model_loading_utils.py
+++ b/src/diffusers/models/model_loading_utils.py
@@ -127,7 +127,7 @@ def _fetch_remapped_cls_from_config(config, old_class):
         return old_class
 
 
-def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[str] = None):
+def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[str] = None, no_mmap: bool = False):
     """
     Reads a checkpoint file, returning properly formatted errors if they arise.
     """
@@ -138,7 +138,10 @@ def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[
     try:
         file_extension = os.path.basename(checkpoint_file).split(".")[-1]
         if file_extension == SAFETENSORS_FILE_EXTENSION:
-            return safetensors.torch.load_file(checkpoint_file, device="cpu")
+            if no_mmap:
+                return safetensors.torch.load(open(checkpoint_file, "rb"), device="cpu")
+            else:
+                return safetensors.torch.load_file(checkpoint_file, device="cpu")
         else:
             weights_only_kwarg = {"weights_only": True} if is_torch_version(">=", "1.13") else {}
             return torch.load(

From a6b4d8fdd3ac9856715c45deb32888f352d57f49 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Tue, 17 Dec 2024 02:27:24 +0000
Subject: [PATCH 02/23] Fix arg parsing.

---
 src/diffusers/loaders/single_file.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 149be4ebcc04..1d6c6c570bff 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -357,6 +357,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         local_files_only = kwargs.pop("local_files_only", False)
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
+        no_mmap = kwargs.pop("no_mmap", False)
 
         is_legacy_loading = False
 

From 3cf01bf7e603ff2f12e9f2c73352bd8d250cfb32 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 16:11:11 +0000
Subject: [PATCH 03/23] Update another method to force no mmap.

---
 src/diffusers/loaders/single_file.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 1d6c6c570bff..666d9e8be139 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -106,6 +106,7 @@ def load_single_file_sub_model(
             subfolder=name,
             torch_dtype=torch_dtype,
             local_files_only=local_files_only,
+            no_mmap=True,
             **kwargs,
         )
 

From 2e082427897eda5f62ae9097b8e0446a7c5cd70d Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 16:31:35 +0000
Subject: [PATCH 04/23] logging

---
 src/diffusers/loaders/single_file.py | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 666d9e8be139..de3c1d28f16e 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -99,6 +99,7 @@ def load_single_file_sub_model(
         if original_config:
             cached_model_config_path = None
 
+        logger.warning("dhipke1")
         loaded_sub_model = load_method(
             pretrained_model_link_or_path_or_dict=checkpoint,
             original_config=original_config,
@@ -111,6 +112,7 @@ def load_single_file_sub_model(
         )
 
     elif is_transformers_model and is_clip_model_in_single_file(class_obj, checkpoint):
+        logger.warning("dhipke2")
         loaded_sub_model = create_diffusers_clip_model_from_ldm(
             class_obj,
             checkpoint=checkpoint,
@@ -142,6 +144,8 @@ def load_single_file_sub_model(
         )
 
     else:
+        logger.warning("dhipke3")
+
         if not hasattr(class_obj, "from_pretrained"):
             raise ValueError(
                 (
@@ -359,6 +363,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
         no_mmap = kwargs.pop("no_mmap", False)
+        logger.warning("no_mmap: " + no_mmap)
 
         is_legacy_loading = False
 

From bcca53b851b6dc371adbe5fe030b329db1b123d9 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 16:46:07 +0000
Subject: [PATCH 05/23] logging2

---
 src/diffusers/loaders/single_file.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index de3c1d28f16e..099462182f19 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -363,7 +363,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
         no_mmap = kwargs.pop("no_mmap", False)
-        logger.warning("no_mmap: " + no_mmap)
+        logger.warning("no_mmap: " + str(no_mmap))
 
         is_legacy_loading = False
 

From c895d8615a7318f7070d5c87781feaf7ec6b9918 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 17:14:09 +0000
Subject: [PATCH 06/23] propagate no_mmap

---
 src/diffusers/loaders/single_file_model.py | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index 78ce47273d8f..540d57dc87df 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -170,6 +170,8 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             revision (`str`, *optional*, defaults to `"main"`):
                 The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                 allowed by Git.
+            no_mmap ('bool', *optional*, defaults to 'False'):
+                Whether to use mmap when loading the model.
             kwargs (remaining dictionary of keyword arguments, *optional*):
                 Can be used to overwrite load and saveable variables (for example the pipeline components of the
                 specific pipeline class). The overwritten components are directly passed to the pipelines `__init__`
@@ -214,6 +216,8 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
         subfolder = kwargs.pop("subfolder", None)
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
+        no_mmap = kwargs.pop("no_mmap", False)
+        logger.warning("no_mmap2: " + str(no_mmap))
 
         if isinstance(pretrained_model_link_or_path_or_dict, dict):
             checkpoint = pretrained_model_link_or_path_or_dict
@@ -226,6 +230,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
                 cache_dir=cache_dir,
                 local_files_only=local_files_only,
                 revision=revision,
+                no_mmap=no_mmap,
             )
 
         mapping_functions = SINGLE_FILE_LOADABLE_CLASSES[mapping_class_name]

From c081e0bddf9570f28c422bd017d063d0e89d707c Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 17:28:16 +0000
Subject: [PATCH 07/23] logging3

---
 src/diffusers/models/model_loading_utils.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/diffusers/models/model_loading_utils.py b/src/diffusers/models/model_loading_utils.py
index a4d5c082e433..deee275b9238 100644
--- a/src/diffusers/models/model_loading_utils.py
+++ b/src/diffusers/models/model_loading_utils.py
@@ -139,8 +139,10 @@ def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[
         file_extension = os.path.basename(checkpoint_file).split(".")[-1]
         if file_extension == SAFETENSORS_FILE_EXTENSION:
             if no_mmap:
+                logger.warning("fast load: " + str(checkpoint_file))
                 return safetensors.torch.load(open(checkpoint_file, "rb"), device="cpu")
             else:
+                logger.warning("slow load: " + str(checkpoint_file))
                 return safetensors.torch.load_file(checkpoint_file, device="cpu")
         else:
             weights_only_kwarg = {"weights_only": True} if is_torch_version(">=", "1.13") else {}

From 7231c282e0e76a5aac2a9df74c639371e57ade36 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 17:39:20 +0000
Subject: [PATCH 08/23] propagate no_mmap

---
 src/diffusers/loaders/single_file_model.py | 2 +-
 src/diffusers/loaders/single_file_utils.py | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index 540d57dc87df..0c3208b463f4 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -318,7 +318,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             unexpected_keys = load_model_dict_into_meta(model, diffusers_format_checkpoint, dtype=torch_dtype)
 
         else:
-            _, unexpected_keys = model.load_state_dict(diffusers_format_checkpoint, strict=False)
+            _, unexpected_keys = model.load_state_dict(diffusers_format_checkpoint, strict=False, no_mmap=no_mmap)
 
         if model._keys_to_ignore_on_load_unexpected is not None:
             for pat in model._keys_to_ignore_on_load_unexpected:
diff --git a/src/diffusers/loaders/single_file_utils.py b/src/diffusers/loaders/single_file_utils.py
index 24fb3452986d..ffee972dd04a 100644
--- a/src/diffusers/loaders/single_file_utils.py
+++ b/src/diffusers/loaders/single_file_utils.py
@@ -382,7 +382,8 @@ def load_single_file_checkpoint(
             revision=revision,
         )
 
-    checkpoint = load_state_dict(pretrained_model_link_or_path, no_mmap)
+    logger.warn("before load_state_dict. no_mmap: " + str(no_mmap))
+    checkpoint = load_state_dict(pretrained_model_link_or_path, no_mmap=no_mmap)
 
     # some checkpoints contain the model state dict under a "state_dict" key
     while "state_dict" in checkpoint:

From 0c472b2fada2815bfa8b9c381f070170732e8590 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 17:51:37 +0000
Subject: [PATCH 09/23] logging4

---
 src/diffusers/loaders/single_file_model.py | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index 0c3208b463f4..bfd495317a70 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -234,8 +234,9 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             )
 
         mapping_functions = SINGLE_FILE_LOADABLE_CLASSES[mapping_class_name]
-
+        logger.warning(mapping_functions)
         checkpoint_mapping_fn = mapping_functions["checkpoint_mapping_fn"]
+        logger.warning(checkpoint_mapping_fn)
         if original_config is not None:
             if "config_mapping_fn" in mapping_functions:
                 config_mapping_fn = mapping_functions["config_mapping_fn"]
@@ -255,6 +256,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
                 # If original_config is a URL or filepath fetch the original_config dict
                 original_config = fetch_original_config(original_config, local_files_only=local_files_only)
 
+            logger.warn("hi1")
             config_mapping_kwargs = _get_mapping_function_kwargs(config_mapping_fn, **kwargs)
             diffusers_model_config = config_mapping_fn(
                 original_config=original_config, checkpoint=checkpoint, **config_mapping_kwargs
@@ -312,14 +314,17 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
 
         ctx = init_empty_weights if is_accelerate_available() else nullcontext
         with ctx():
+            logger.warn("hi2")
             model = cls.from_config(diffusers_model_config)
 
+        logger.warn("hi3")
         if is_accelerate_available():
             unexpected_keys = load_model_dict_into_meta(model, diffusers_format_checkpoint, dtype=torch_dtype)
 
         else:
             _, unexpected_keys = model.load_state_dict(diffusers_format_checkpoint, strict=False, no_mmap=no_mmap)
 
+        logger.warn("hi4")
         if model._keys_to_ignore_on_load_unexpected is not None:
             for pat in model._keys_to_ignore_on_load_unexpected:
                 unexpected_keys = [k for k in unexpected_keys if re.search(pat, k) is None]

From c4d4d60cd28ff4539d6cecef1d04ae479b6c605d Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Wed, 18 Dec 2024 17:56:32 +0000
Subject: [PATCH 10/23] fix open call

---
 src/diffusers/models/model_loading_utils.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/diffusers/models/model_loading_utils.py b/src/diffusers/models/model_loading_utils.py
index deee275b9238..06a60d4cc839 100644
--- a/src/diffusers/models/model_loading_utils.py
+++ b/src/diffusers/models/model_loading_utils.py
@@ -140,7 +140,7 @@ def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[
         if file_extension == SAFETENSORS_FILE_EXTENSION:
             if no_mmap:
                 logger.warning("fast load: " + str(checkpoint_file))
-                return safetensors.torch.load(open(checkpoint_file, "rb"), device="cpu")
+                return safetensors.torch.load(open(checkpoint_file, "rb").read())
             else:
                 logger.warning("slow load: " + str(checkpoint_file))
                 return safetensors.torch.load_file(checkpoint_file, device="cpu")

From 4f84222d1eb3776e12ffdaf44169bd4e9c36b6ca Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Thu, 19 Dec 2024 16:59:04 +0000
Subject: [PATCH 11/23] clean up logging

---
 src/diffusers/loaders/single_file.py       | 5 -----
 src/diffusers/loaders/single_file_model.py | 7 -------
 src/diffusers/loaders/single_file_utils.py | 1 -
 3 files changed, 13 deletions(-)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 099462182f19..666d9e8be139 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -99,7 +99,6 @@ def load_single_file_sub_model(
         if original_config:
             cached_model_config_path = None
 
-        logger.warning("dhipke1")
         loaded_sub_model = load_method(
             pretrained_model_link_or_path_or_dict=checkpoint,
             original_config=original_config,
@@ -112,7 +111,6 @@ def load_single_file_sub_model(
         )
 
     elif is_transformers_model and is_clip_model_in_single_file(class_obj, checkpoint):
-        logger.warning("dhipke2")
         loaded_sub_model = create_diffusers_clip_model_from_ldm(
             class_obj,
             checkpoint=checkpoint,
@@ -144,8 +142,6 @@ def load_single_file_sub_model(
         )
 
     else:
-        logger.warning("dhipke3")
-
         if not hasattr(class_obj, "from_pretrained"):
             raise ValueError(
                 (
@@ -363,7 +359,6 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
         no_mmap = kwargs.pop("no_mmap", False)
-        logger.warning("no_mmap: " + str(no_mmap))
 
         is_legacy_loading = False
 
diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index bfd495317a70..eb72d549f30e 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -217,7 +217,6 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
         no_mmap = kwargs.pop("no_mmap", False)
-        logger.warning("no_mmap2: " + str(no_mmap))
 
         if isinstance(pretrained_model_link_or_path_or_dict, dict):
             checkpoint = pretrained_model_link_or_path_or_dict
@@ -234,9 +233,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             )
 
         mapping_functions = SINGLE_FILE_LOADABLE_CLASSES[mapping_class_name]
-        logger.warning(mapping_functions)
         checkpoint_mapping_fn = mapping_functions["checkpoint_mapping_fn"]
-        logger.warning(checkpoint_mapping_fn)
         if original_config is not None:
             if "config_mapping_fn" in mapping_functions:
                 config_mapping_fn = mapping_functions["config_mapping_fn"]
@@ -256,7 +253,6 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
                 # If original_config is a URL or filepath fetch the original_config dict
                 original_config = fetch_original_config(original_config, local_files_only=local_files_only)
 
-            logger.warn("hi1")
             config_mapping_kwargs = _get_mapping_function_kwargs(config_mapping_fn, **kwargs)
             diffusers_model_config = config_mapping_fn(
                 original_config=original_config, checkpoint=checkpoint, **config_mapping_kwargs
@@ -314,17 +310,14 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
 
         ctx = init_empty_weights if is_accelerate_available() else nullcontext
         with ctx():
-            logger.warn("hi2")
             model = cls.from_config(diffusers_model_config)
 
-        logger.warn("hi3")
         if is_accelerate_available():
             unexpected_keys = load_model_dict_into_meta(model, diffusers_format_checkpoint, dtype=torch_dtype)
 
         else:
             _, unexpected_keys = model.load_state_dict(diffusers_format_checkpoint, strict=False, no_mmap=no_mmap)
 
-        logger.warn("hi4")
         if model._keys_to_ignore_on_load_unexpected is not None:
             for pat in model._keys_to_ignore_on_load_unexpected:
                 unexpected_keys = [k for k in unexpected_keys if re.search(pat, k) is None]
diff --git a/src/diffusers/loaders/single_file_utils.py b/src/diffusers/loaders/single_file_utils.py
index ffee972dd04a..4586e7d07b79 100644
--- a/src/diffusers/loaders/single_file_utils.py
+++ b/src/diffusers/loaders/single_file_utils.py
@@ -382,7 +382,6 @@ def load_single_file_checkpoint(
             revision=revision,
         )
 
-    logger.warn("before load_state_dict. no_mmap: " + str(no_mmap))
     checkpoint = load_state_dict(pretrained_model_link_or_path, no_mmap=no_mmap)
 
     # some checkpoints contain the model state dict under a "state_dict" key

From 5fab6d18b953b22785050bfe13761664b9e602bd Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Thu, 19 Dec 2024 17:14:48 +0000
Subject: [PATCH 12/23] cleanup

---
 src/diffusers/loaders/single_file.py       | 6 ++++--
 src/diffusers/loaders/single_file_model.py | 4 +++-
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 666d9e8be139..54d61023e9cc 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -106,7 +106,7 @@ def load_single_file_sub_model(
             subfolder=name,
             torch_dtype=torch_dtype,
             local_files_only=local_files_only,
-            no_mmap=True,
+            no_mmap=no_mmap,
             **kwargs,
         )
 
@@ -310,7 +310,8 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
                     - A path to a *directory* (for example `./my_pipeline_directory/`) containing the pipeline
                       component configs in Diffusers format.
             no_mmap ('bool', *optional*, defaults to 'False'):
-                Whether to use mmap when loading the model.
+                Whether to disable mmap when loading the model. This option can perform better when the model is on
+                a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
             kwargs (remaining dictionary of keyword arguments, *optional*):
                 Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline
                 class). The overwritten components are passed directly to the pipelines `__init__` method. See example
@@ -509,6 +510,7 @@ def load_module(name, value):
                         original_config=original_config,
                         local_files_only=local_files_only,
                         is_legacy_loading=is_legacy_loading,
+                        no_mmap=no_mmap,
                         **kwargs,
                     )
                 except SingleFileComponentError as e:
diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index eb72d549f30e..f392b0492bb1 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -171,7 +171,8 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
                 The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                 allowed by Git.
             no_mmap ('bool', *optional*, defaults to 'False'):
-                Whether to use mmap when loading the model.
+                Whether to disable mmap when loading the model. This option can perform better when the model is on
+                a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
             kwargs (remaining dictionary of keyword arguments, *optional*):
                 Can be used to overwrite load and saveable variables (for example the pipeline components of the
                 specific pipeline class). The overwritten components are directly passed to the pipelines `__init__`
@@ -233,6 +234,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             )
 
         mapping_functions = SINGLE_FILE_LOADABLE_CLASSES[mapping_class_name]
+
         checkpoint_mapping_fn = mapping_functions["checkpoint_mapping_fn"]
         if original_config is not None:
             if "config_mapping_fn" in mapping_functions:

From 1d8cf69ac8e173dde9a3c3f9bf428406748acd03 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Thu, 19 Dec 2024 17:20:19 +0000
Subject: [PATCH 13/23] fix missing arg

---
 src/diffusers/loaders/single_file.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 54d61023e9cc..7b614e97f6fa 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -60,6 +60,7 @@ def load_single_file_sub_model(
     local_files_only=False,
     torch_dtype=None,
     is_legacy_loading=False,
+    no_mmap=False,
     **kwargs,
 ):
     if is_pipeline_module:

From 5ef288fdb3e1128fc1e5c578c59610d958401c7e Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Thu, 19 Dec 2024 17:25:40 +0000
Subject: [PATCH 14/23] update logging and comments

---
 src/diffusers/loaders/single_file.py        | 4 ++--
 src/diffusers/loaders/single_file_model.py  | 4 ++--
 src/diffusers/models/model_loading_utils.py | 2 --
 3 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 7b614e97f6fa..b277eb1b5e2b 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -311,8 +311,8 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
                     - A path to a *directory* (for example `./my_pipeline_directory/`) containing the pipeline
                       component configs in Diffusers format.
             no_mmap ('bool', *optional*, defaults to 'False'):
-                Whether to disable mmap when loading the model. This option can perform better when the model is on
-                a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
+                Whether to disable mmap when loading a Safetensors model. This option can perform better when the model
+                is on a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
             kwargs (remaining dictionary of keyword arguments, *optional*):
                 Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline
                 class). The overwritten components are passed directly to the pipelines `__init__` method. See example
diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index f392b0492bb1..a2eab5e6b60f 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -171,8 +171,8 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
                 The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                 allowed by Git.
             no_mmap ('bool', *optional*, defaults to 'False'):
-                Whether to disable mmap when loading the model. This option can perform better when the model is on
-                a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
+                Whether to disable mmap when loading a Safetensors model. This option can perform better when the model
+                is on a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
             kwargs (remaining dictionary of keyword arguments, *optional*):
                 Can be used to overwrite load and saveable variables (for example the pipeline components of the
                 specific pipeline class). The overwritten components are directly passed to the pipelines `__init__`
diff --git a/src/diffusers/models/model_loading_utils.py b/src/diffusers/models/model_loading_utils.py
index 06a60d4cc839..c9fa90252527 100644
--- a/src/diffusers/models/model_loading_utils.py
+++ b/src/diffusers/models/model_loading_utils.py
@@ -139,10 +139,8 @@ def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[
         file_extension = os.path.basename(checkpoint_file).split(".")[-1]
         if file_extension == SAFETENSORS_FILE_EXTENSION:
             if no_mmap:
-                logger.warning("fast load: " + str(checkpoint_file))
                 return safetensors.torch.load(open(checkpoint_file, "rb").read())
             else:
-                logger.warning("slow load: " + str(checkpoint_file))
                 return safetensors.torch.load_file(checkpoint_file, device="cpu")
         else:
             weights_only_kwarg = {"weights_only": True} if is_torch_version(">=", "1.13") else {}

From f80644d1c052852a2940014f49a5e5ed968a9e6e Mon Sep 17 00:00:00 2001
From: Daniel Hipke <dhipke@google.com>
Date: Fri, 20 Dec 2024 21:21:46 +0000
Subject: [PATCH 15/23] Rename to disable_mmap and update other references.

---
 src/diffusers/loaders/single_file.py        | 12 ++++++------
 src/diffusers/loaders/single_file_model.py  | 10 ++++++----
 src/diffusers/loaders/single_file_utils.py  |  4 ++--
 src/diffusers/models/model_loading_utils.py |  6 ++++--
 src/diffusers/models/modeling_utils.py      |  8 ++++++--
 5 files changed, 24 insertions(+), 16 deletions(-)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index b277eb1b5e2b..68a5c21ca732 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -60,7 +60,7 @@ def load_single_file_sub_model(
     local_files_only=False,
     torch_dtype=None,
     is_legacy_loading=False,
-    no_mmap=False,
+    disable_mmap=False,
     **kwargs,
 ):
     if is_pipeline_module:
@@ -107,7 +107,7 @@ def load_single_file_sub_model(
             subfolder=name,
             torch_dtype=torch_dtype,
             local_files_only=local_files_only,
-            no_mmap=no_mmap,
+            disable_mmap=disable_mmap,
             **kwargs,
         )
 
@@ -310,7 +310,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
                       hosted on the Hub.
                     - A path to a *directory* (for example `./my_pipeline_directory/`) containing the pipeline
                       component configs in Diffusers format.
-            no_mmap ('bool', *optional*, defaults to 'False'):
+            disable_mmap ('bool', *optional*, defaults to 'False'):
                 Whether to disable mmap when loading a Safetensors model. This option can perform better when the model
                 is on a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
             kwargs (remaining dictionary of keyword arguments, *optional*):
@@ -360,7 +360,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
         local_files_only = kwargs.pop("local_files_only", False)
         revision = kwargs.pop("revision", None)
         torch_dtype = kwargs.pop("torch_dtype", None)
-        no_mmap = kwargs.pop("no_mmap", False)
+        disable_mmap = kwargs.pop("disable_mmap", False)
 
         is_legacy_loading = False
 
@@ -389,7 +389,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
             cache_dir=cache_dir,
             local_files_only=local_files_only,
             revision=revision,
-            no_mmap=no_mmap,
+            disable_mmap=disable_mmap,
         )
 
         if config is None:
@@ -511,7 +511,7 @@ def load_module(name, value):
                         original_config=original_config,
                         local_files_only=local_files_only,
                         is_legacy_loading=is_legacy_loading,
-                        no_mmap=no_mmap,
+                        disable_mmap=disable_mmap,
                         **kwargs,
                     )
                 except SingleFileComponentError as e:
diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index 9abc730722c4..fe9f2e14ceeb 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -177,7 +177,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             revision (`str`, *optional*, defaults to `"main"`):
                 The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                 allowed by Git.
-            no_mmap ('bool', *optional*, defaults to 'False'):
+            disable_mmap ('bool', *optional*, defaults to 'False'):
                 Whether to disable mmap when loading a Safetensors model. This option can perform better when the model
                 is on a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
             kwargs (remaining dictionary of keyword arguments, *optional*):
@@ -226,7 +226,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
         torch_dtype = kwargs.pop("torch_dtype", None)
         quantization_config = kwargs.pop("quantization_config", None)
         device = kwargs.pop("device", None)
-        no_mmap = kwargs.pop("no_mmap", False)
+        disable_mmap = kwargs.pop("disable_mmap", False)
 
         if isinstance(pretrained_model_link_or_path_or_dict, dict):
             checkpoint = pretrained_model_link_or_path_or_dict
@@ -239,7 +239,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
                 cache_dir=cache_dir,
                 local_files_only=local_files_only,
                 revision=revision,
-                no_mmap=no_mmap,
+                disable_mmap=disable_mmap,
             )
         if quantization_config is not None:
             hf_quantizer = DiffusersAutoQuantizer.from_config(quantization_config)
@@ -361,7 +361,9 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             )
 
         else:
-            _, unexpected_keys = model.load_state_dict(diffusers_format_checkpoint, strict=False, no_mmap=no_mmap)
+            _, unexpected_keys = model.load_state_dict(
+                diffusers_format_checkpoint, strict=False, disable_mmap=disable_mmap
+            )
 
         if model._keys_to_ignore_on_load_unexpected is not None:
             for pat in model._keys_to_ignore_on_load_unexpected:
diff --git a/src/diffusers/loaders/single_file_utils.py b/src/diffusers/loaders/single_file_utils.py
index 46791a9c2f32..a21686b7b388 100644
--- a/src/diffusers/loaders/single_file_utils.py
+++ b/src/diffusers/loaders/single_file_utils.py
@@ -375,7 +375,7 @@ def load_single_file_checkpoint(
     cache_dir=None,
     local_files_only=None,
     revision=None,
-    no_mmap=False,
+    disable_mmap=False,
 ):
     if os.path.isfile(pretrained_model_link_or_path):
         pretrained_model_link_or_path = pretrained_model_link_or_path
@@ -393,7 +393,7 @@ def load_single_file_checkpoint(
             revision=revision,
         )
 
-    checkpoint = load_state_dict(pretrained_model_link_or_path, no_mmap=no_mmap)
+    checkpoint = load_state_dict(pretrained_model_link_or_path, disable_mmap=disable_mmap)
 
     # some checkpoints contain the model state dict under a "state_dict" key
     while "state_dict" in checkpoint:
diff --git a/src/diffusers/models/model_loading_utils.py b/src/diffusers/models/model_loading_utils.py
index 6e9ff8261761..cc8e81a0b90c 100644
--- a/src/diffusers/models/model_loading_utils.py
+++ b/src/diffusers/models/model_loading_utils.py
@@ -131,7 +131,9 @@ def _fetch_remapped_cls_from_config(config, old_class):
         return old_class
 
 
-def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[str] = None, no_mmap: bool = False):
+def load_state_dict(
+    checkpoint_file: Union[str, os.PathLike], variant: Optional[str] = None, disable_mmap: bool = False
+):
     """
     Reads a checkpoint file, returning properly formatted errors if they arise.
     """
@@ -142,7 +144,7 @@ def load_state_dict(checkpoint_file: Union[str, os.PathLike], variant: Optional[
     try:
         file_extension = os.path.basename(checkpoint_file).split(".")[-1]
         if file_extension == SAFETENSORS_FILE_EXTENSION:
-            if no_mmap:
+            if disable_mmap:
                 return safetensors.torch.load(open(checkpoint_file, "rb").read())
             else:
                 return safetensors.torch.load_file(checkpoint_file, device="cpu")
diff --git a/src/diffusers/models/modeling_utils.py b/src/diffusers/models/modeling_utils.py
index 872d4d73d41f..e16f44bc73d3 100644
--- a/src/diffusers/models/modeling_utils.py
+++ b/src/diffusers/models/modeling_utils.py
@@ -541,6 +541,9 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
                 If set to `None`, the `safetensors` weights are downloaded if they're available **and** if the
                 `safetensors` library is installed. If set to `True`, the model is forcibly loaded from `safetensors`
                 weights. If set to `False`, `safetensors` weights are not loaded.
+            disable_mmap ('bool', *optional*, defaults to 'False'):
+                Whether to disable mmap when loading a Safetensors model. This option can perform better when the model
+                is on a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
 
         <Tip>
 
@@ -586,6 +589,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
         variant = kwargs.pop("variant", None)
         use_safetensors = kwargs.pop("use_safetensors", None)
         quantization_config = kwargs.pop("quantization_config", None)
+        disable_mmap = kwargs.pop("disable_mmap", False)
 
         allow_pickle = False
         if use_safetensors is None:
@@ -865,7 +869,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
                     # TODO (sayakpaul,  SunMarc): remove this after model loading refactor
                     else:
                         param_device = torch.device(torch.cuda.current_device())
-                    state_dict = load_state_dict(model_file, variant=variant)
+                    state_dict = load_state_dict(model_file, variant=variant, disable_mmap=disable_mmap)
                     model._convert_deprecated_attention_blocks(state_dict)
 
                     # move the params from meta device to cpu
@@ -965,7 +969,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
             else:
                 model = cls.from_config(config, **unused_kwargs)
 
-                state_dict = load_state_dict(model_file, variant=variant)
+                state_dict = load_state_dict(model_file, variant=variant, disable_mmap=disable_mmap)
                 model._convert_deprecated_attention_blocks(state_dict)
 
                 model, missing_keys, unexpected_keys, mismatched_keys, error_msgs = cls._load_pretrained_model(

From ffe5abae5e8287f176139cab7287dc9ef9edc48c Mon Sep 17 00:00:00 2001
From: Sayak Paul <spsayakpaul@gmail.com>
Date: Fri, 20 Dec 2024 18:32:22 +0530
Subject: [PATCH 16/23] [Docs] Update ltx_video.md to remove generator from
 `from_pretrained()` (#10316)

Update ltx_video.md to remove generator from `from_pretrained()`
---
 docs/source/en/api/pipelines/ltx_video.md | 1 -
 1 file changed, 1 deletion(-)

diff --git a/docs/source/en/api/pipelines/ltx_video.md b/docs/source/en/api/pipelines/ltx_video.md
index 211cd3007d1e..a925b848706e 100644
--- a/docs/source/en/api/pipelines/ltx_video.md
+++ b/docs/source/en/api/pipelines/ltx_video.md
@@ -79,7 +79,6 @@ transformer = LTXVideoTransformer3DModel.from_single_file(
 pipe = LTXPipeline.from_pretrained(
     "Lightricks/LTX-Video",
     transformer=transformer,
-    generator=torch.manual_seed(0),
     torch_dtype=torch.bfloat16,
 )
 pipe.enable_model_cpu_offload()

From 3fc4a422e54aa79a47d6ab1689cda9dd2a00131e Mon Sep 17 00:00:00 2001
From: Leojc <liao_junchao@outlook.com>
Date: Fri, 20 Dec 2024 23:22:32 +0800
Subject: [PATCH 17/23] docs: fix a mistake in docstring (#10319)

Update pipeline_hunyuan_video.py

docs: fix a mistake
---
 src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py b/src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py
index 4423ccf97932..6e0541e938ba 100644
--- a/src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py
+++ b/src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py
@@ -143,7 +143,7 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
     Args:
         text_encoder ([`LlamaModel`]):
             [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
-        tokenizer_2 (`LlamaTokenizer`):
+        tokenizer (`LlamaTokenizer`):
             Tokenizer from [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
         transformer ([`HunyuanVideoTransformer3DModel`]):
             Conditional Transformer to denoise the encoded image latents.

From 9e887b478aebcadbfd242c2f06d329d60f3e262c Mon Sep 17 00:00:00 2001
From: Aditya Raj <syntaxticsugr@gmail.com>
Date: Fri, 20 Dec 2024 20:59:58 +0530
Subject: [PATCH 18/23] [BUG FIX] [Stable Audio Pipeline] Resolve
 torch.Tensor.new_zeros() TypeError in function prepare_latents caused by
 audio_vae_length (#10306)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

[BUG FIX] [Stable Audio Pipeline] TypeError: new_zeros(): argument 'size' failed to unpack the object at pos 3 with error "type must be tuple of ints,but got float"

torch.Tensor.new_zeros() takes a single argument size (int...) – a list, tuple, or torch.Size of integers defining the shape of the output tensor.

in function prepare_latents:
audio_vae_length = self.transformer.config.sample_size * self.vae.hop_length
audio_shape = (batch_size // num_waveforms_per_prompt, audio_channels, audio_vae_length)
...
audio = initial_audio_waveforms.new_zeros(audio_shape)

audio_vae_length evaluates to float because self.transformer.config.sample_size returns a float

Co-authored-by: hlky <hlky@hlky.ac>
---
 src/diffusers/pipelines/stable_audio/pipeline_stable_audio.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/diffusers/pipelines/stable_audio/pipeline_stable_audio.py b/src/diffusers/pipelines/stable_audio/pipeline_stable_audio.py
index cef63cf7e63d..5d773b614a5c 100644
--- a/src/diffusers/pipelines/stable_audio/pipeline_stable_audio.py
+++ b/src/diffusers/pipelines/stable_audio/pipeline_stable_audio.py
@@ -446,7 +446,7 @@ def prepare_latents(
                     f"`initial_audio_waveforms` must be of shape `(batch_size, num_channels, audio_length)` or `(batch_size, audio_length)` but has `{initial_audio_waveforms.ndim}` dimensions"
                 )
 
-            audio_vae_length = self.transformer.config.sample_size * self.vae.hop_length
+            audio_vae_length = int(self.transformer.config.sample_size) * self.vae.hop_length
             audio_shape = (batch_size // num_waveforms_per_prompt, audio_channels, audio_vae_length)
 
             # check num_channels

From dbbcd0f343c7b87c444ea757b5e100c674ca5123 Mon Sep 17 00:00:00 2001
From: Steven Liu <59462357+stevhliu@users.noreply.github.com>
Date: Fri, 20 Dec 2024 08:30:21 -0800
Subject: [PATCH 19/23] [docs] Fix quantization links (#10323)

Update overview.md
---
 docs/source/en/quantization/overview.md | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/docs/source/en/quantization/overview.md b/docs/source/en/quantization/overview.md
index 3eef5238f1ce..794098e210a6 100644
--- a/docs/source/en/quantization/overview.md
+++ b/docs/source/en/quantization/overview.md
@@ -33,8 +33,8 @@ If you are new to the quantization field, we recommend you to check out these be
 ## When to use what?
 
 Diffusers currently supports the following quantization methods.
-- [BitsandBytes](./bitsandbytes.md)
-- [TorchAO](./torchao.md)
-- [GGUF](./gguf.md)
+- [BitsandBytes](./bitsandbytes)
+- [TorchAO](./torchao)
+- [GGUF](./gguf)
 
 [This resource](https://huggingface.co/docs/transformers/main/en/quantization/overview#when-to-use-what) provides a good overview of the pros and cons of different quantization techniques.

From dfebda27c25b5b650674338d721524c2fd227e1b Mon Sep 17 00:00:00 2001
From: Junsong Chen <cjs1020440147@icloud.com>
Date: Sat, 21 Dec 2024 01:21:34 +0800
Subject: [PATCH 20/23] [Sana]add 2K related model for Sana (#10322)

add 2K related model for Sana
---
 scripts/convert_sana_to_diffusers.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/scripts/convert_sana_to_diffusers.py b/scripts/convert_sana_to_diffusers.py
index c1045a98a51a..dc553681678b 100644
--- a/scripts/convert_sana_to_diffusers.py
+++ b/scripts/convert_sana_to_diffusers.py
@@ -25,6 +25,7 @@
 CTX = init_empty_weights if is_accelerate_available else nullcontext
 
 ckpt_ids = [
+    "Efficient-Large-Model/Sana_1600M_2Kpx_BF16/checkpoints/Sana_1600M_2Kpx_BF16.pth",
     "Efficient-Large-Model/Sana_1600M_1024px_MultiLing/checkpoints/Sana_1600M_1024px_MultiLing.pth",
     "Efficient-Large-Model/Sana_1600M_1024px_BF16/checkpoints/Sana_1600M_1024px_BF16.pth",
     "Efficient-Large-Model/Sana_1600M_512px_MultiLing/checkpoints/Sana_1600M_512px_MultiLing.pth",
@@ -265,9 +266,9 @@ def main(args):
         "--image_size",
         default=1024,
         type=int,
-        choices=[512, 1024],
+        choices=[512, 1024, 2048],
         required=False,
-        help="Image size of pretrained model, 512 or 1024.",
+        help="Image size of pretrained model, 512, 1024 or 2048.",
     )
     parser.add_argument(
         "--model_type", default="SanaMS_1600M_P1_D20", type=str, choices=["SanaMS_1600M_P1_D20", "SanaMS_600M_P1_D28"]

From 6720c51ae78f7231c315b4ab04fbf5704d3c30a7 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <danhipke@users.noreply.github.com>
Date: Sun, 22 Dec 2024 22:59:26 -0800
Subject: [PATCH 21/23] Update src/diffusers/loaders/single_file_model.py

Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
---
 src/diffusers/loaders/single_file_model.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index fe9f2e14ceeb..70178fde6cbf 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -362,7 +362,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
 
         else:
             _, unexpected_keys = model.load_state_dict(
-                diffusers_format_checkpoint, strict=False, disable_mmap=disable_mmap
+                diffusers_format_checkpoint, strict=False
             )
 
         if model._keys_to_ignore_on_load_unexpected is not None:

From 2926158e7ff23f13e5992755e89e284abb239819 Mon Sep 17 00:00:00 2001
From: Daniel Hipke <danhipke@users.noreply.github.com>
Date: Sun, 22 Dec 2024 22:59:42 -0800
Subject: [PATCH 22/23] Update src/diffusers/loaders/single_file.py

Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
---
 src/diffusers/loaders/single_file.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/diffusers/loaders/single_file.py b/src/diffusers/loaders/single_file.py
index 68a5c21ca732..71161cc58529 100644
--- a/src/diffusers/loaders/single_file.py
+++ b/src/diffusers/loaders/single_file.py
@@ -312,7 +312,7 @@ def from_single_file(cls, pretrained_model_link_or_path, **kwargs):
                       component configs in Diffusers format.
             disable_mmap ('bool', *optional*, defaults to 'False'):
                 Whether to disable mmap when loading a Safetensors model. This option can perform better when the model
-                is on a network mount or hard drive, which may not handle the seeky-ness of mmap very well.
+                is on a network mount or hard drive.
             kwargs (remaining dictionary of keyword arguments, *optional*):
                 Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline
                 class). The overwritten components are passed directly to the pipelines `__init__` method. See example

From 22b33708a152f101075f4c15e538e2bfab5825fd Mon Sep 17 00:00:00 2001
From: DN6 <dhruv.nair@gmail.com>
Date: Fri, 10 Jan 2025 13:57:09 +0530
Subject: [PATCH 23/23] make style

---
 src/diffusers/loaders/single_file_model.py | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/src/diffusers/loaders/single_file_model.py b/src/diffusers/loaders/single_file_model.py
index 4d294b262a42..0c998bab5e0f 100644
--- a/src/diffusers/loaders/single_file_model.py
+++ b/src/diffusers/loaders/single_file_model.py
@@ -367,9 +367,7 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             )
 
         else:
-            _, unexpected_keys = model.load_state_dict(
-                diffusers_format_checkpoint, strict=False
-            )
+            _, unexpected_keys = model.load_state_dict(diffusers_format_checkpoint, strict=False)
 
         if model._keys_to_ignore_on_load_unexpected is not None:
             for pat in model._keys_to_ignore_on_load_unexpected: