diffusers 0.35.1__py3-none-any.whl → 0.35.2__py3-none-any.whl

diffusers/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.35.1"
+__version__ = "0.35.2"
 
 from typing import TYPE_CHECKING
 

diffusers/models/attention_dispatch.py

@@ -110,6 +110,27 @@ if _CAN_USE_XFORMERS_ATTN:
 else:
     xops = None
 
+# Version guard for PyTorch compatibility - custom_op was added in PyTorch 2.4
+if torch.__version__ >= "2.4.0":
+    _custom_op = torch.library.custom_op
+    _register_fake = torch.library.register_fake
+else:
+
+    def custom_op_no_op(name, fn=None, /, *, mutates_args, device_types=None, schema=None):
+        def wrap(func):
+            return func
+
+        return wrap if fn is None else fn
+
+    def register_fake_no_op(op, fn=None, /, *, lib=None, _stacklevel=1):
+        def wrap(func):
+            return func
+
+        return wrap if fn is None else fn
+
+    _custom_op = custom_op_no_op
+    _register_fake = register_fake_no_op
+
 
 logger = get_logger(__name__)  # pylint: disable=invalid-name
 
@@ -473,12 +494,11 @@ def _flex_attention_causal_mask_mod(batch_idx, head_idx, q_idx, kv_idx):
 
 # ===== torch op registrations =====
 # Registrations are required for fullgraph tracing compatibility
-
-
-# TODO: library.custom_op and register_fake probably need version guards?
 # TODO: this is only required because the beta release FA3 does not have it. There is a PR adding
 # this but it was never merged: https://github.com/Dao-AILab/flash-attention/pull/1590
-@torch.library.custom_op("flash_attn_3::_flash_attn_forward", mutates_args=(), device_types="cuda")
+
+
+@_custom_op("flash_attn_3::_flash_attn_forward", mutates_args=(), device_types="cuda")
 def _wrapped_flash_attn_3_original(
     query: torch.Tensor, key: torch.Tensor, value: torch.Tensor
 ) -> Tuple[torch.Tensor, torch.Tensor]:
@@ -487,7 +507,7 @@ def _wrapped_flash_attn_3_original(
     return out, lse
 
 
-@torch.library.register_fake("flash_attn_3::_flash_attn_forward")
+@_register_fake("flash_attn_3::_flash_attn_forward")
 def _(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
     batch_size, seq_len, num_heads, head_dim = query.shape
     lse_shape = (batch_size, seq_len, num_heads)

diffusers/models/transformers/transformer_ltx.py

@@ -350,7 +350,9 @@ class LTXVideoTransformerBlock(nn.Module):
         norm_hidden_states = self.norm1(hidden_states)
 
         num_ada_params = self.scale_shift_table.shape[0]
-        ada_values = self.scale_shift_table[None, None] + temb.reshape(batch_size, temb.size(1), num_ada_params, -1)
+        ada_values = self.scale_shift_table[None, None].to(temb.device) + temb.reshape(
+            batch_size, temb.size(1), num_ada_params, -1
+        )
         shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = ada_values.unbind(dim=2)
         norm_hidden_states = norm_hidden_states * (1 + scale_msa) + shift_msa
 

diffusers/models/transformers/transformer_wan.py

@@ -665,12 +665,12 @@ class WanTransformer3DModel(
         # 5. Output norm, projection & unpatchify
         if temb.ndim == 3:
             # batch_size, seq_len, inner_dim (wan 2.2 ti2v)
-            shift, scale = (self.scale_shift_table.unsqueeze(0) + temb.unsqueeze(2)).chunk(2, dim=2)
+            shift, scale = (self.scale_shift_table.unsqueeze(0).to(temb.device) + temb.unsqueeze(2)).chunk(2, dim=2)
             shift = shift.squeeze(2)
             scale = scale.squeeze(2)
         else:
             # batch_size, inner_dim
-            shift, scale = (self.scale_shift_table + temb.unsqueeze(1)).chunk(2, dim=1)
+            shift, scale = (self.scale_shift_table.to(temb.device) + temb.unsqueeze(1)).chunk(2, dim=1)
 
         # Move the shift and scale tensors to the same device as hidden_states.
         # When using multi-GPU inference via accelerate these will be on the

diffusers/models/transformers/transformer_wan_vace.py

@@ -103,7 +103,7 @@ class WanVACETransformerBlock(nn.Module):
             control_hidden_states = control_hidden_states + hidden_states
 
         shift_msa, scale_msa, gate_msa, c_shift_msa, c_scale_msa, c_gate_msa = (
-            self.scale_shift_table + temb.float()
+            self.scale_shift_table.to(temb.device) + temb.float()
         ).chunk(6, dim=1)
 
         # 1. Self-attention
@@ -359,7 +359,7 @@ class WanVACETransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromO
                     hidden_states = hidden_states + control_hint * scale
 
         # 6. Output norm, projection & unpatchify
-        shift, scale = (self.scale_shift_table + temb.unsqueeze(1)).chunk(2, dim=1)
+        shift, scale = (self.scale_shift_table.to(temb.device) + temb.unsqueeze(1)).chunk(2, dim=1)
 
         # Move the shift and scale tensors to the same device as hidden_states.
         # When using multi-GPU inference via accelerate these will be on the

diffusers/pipelines/pipeline_loading_utils.py

@@ -48,10 +48,12 @@ from .transformers_loading_utils import _load_tokenizer_from_dduf, _load_transfo
 if is_transformers_available():
     import transformers
     from transformers import PreTrainedModel, PreTrainedTokenizerBase
-    from transformers.utils import FLAX_WEIGHTS_NAME as TRANSFORMERS_FLAX_WEIGHTS_NAME
     from transformers.utils import SAFE_WEIGHTS_NAME as TRANSFORMERS_SAFE_WEIGHTS_NAME
     from transformers.utils import WEIGHTS_NAME as TRANSFORMERS_WEIGHTS_NAME
 
+    if is_transformers_version("<=", "4.56.2"):
+        from transformers.utils import FLAX_WEIGHTS_NAME as TRANSFORMERS_FLAX_WEIGHTS_NAME
+
 if is_accelerate_available():
     import accelerate
     from accelerate import dispatch_model
@@ -112,7 +114,9 @@ def is_safetensors_compatible(filenames, passed_components=None, folder_names=No
     ]
 
     if is_transformers_available():
-        weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME, TRANSFORMERS_FLAX_WEIGHTS_NAME]
+        weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME]
+        if is_transformers_version("<=", "4.56.2"):
+            weight_names += [TRANSFORMERS_FLAX_WEIGHTS_NAME]
 
     # model_pytorch, diffusion_model_pytorch, ...
     weight_prefixes = [w.split(".")[0] for w in weight_names]
@@ -191,7 +195,9 @@ def filter_model_files(filenames):
     ]
 
     if is_transformers_available():
-        weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME, TRANSFORMERS_FLAX_WEIGHTS_NAME]
+        weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME]
+        if is_transformers_version("<=", "4.56.2"):
+            weight_names += [TRANSFORMERS_FLAX_WEIGHTS_NAME]
 
     allowed_extensions = [wn.split(".")[-1] for wn in weight_names]
 
@@ -212,7 +218,9 @@ def variant_compatible_siblings(filenames, variant=None, ignore_patterns=None) -
     ]
 
     if is_transformers_available():
-        weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME, TRANSFORMERS_FLAX_WEIGHTS_NAME]
+        weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME]
+        if is_transformers_version("<=", "4.56.2"):
+            weight_names += [TRANSFORMERS_FLAX_WEIGHTS_NAME]
 
     # model_pytorch, diffusion_model_pytorch, ...
     weight_prefixes = [w.split(".")[0] for w in weight_names]
@@ -830,6 +838,9 @@ def load_sub_model(
         else:
             loading_kwargs["low_cpu_mem_usage"] = False
 
+    if is_transformers_model and is_transformers_version(">=", "4.57.0"):
+        loading_kwargs.pop("offload_state_dict")
+
     if (
         quantization_config is not None
         and isinstance(quantization_config, PipelineQuantizationConfig)

{diffusers-0.35.1.dist-info → diffusers-0.35.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: diffusers
-Version: 0.35.1
+Version: 0.35.2
 Summary: State-of-the-art diffusion in PyTorch and JAX.
 Home-page: https://github.com/huggingface/diffusers
 Author: The Hugging Face team (past and future) with the help of all our contributors (https://github.com/huggingface/diffusers/graphs/contributors)

{diffusers-0.35.1.dist-info → diffusers-0.35.2.dist-info}/RECORD

@@ -1,4 +1,4 @@
-diffusers/__init__.py,sha256=ez_SaG7tS8y601bNDEVjOX-_7BXe7KdUrG9PCsQ6kY0,50085
+diffusers/__init__.py,sha256=otVY6ZdoA1vJBd-2_zRA-_0NVYvhbKzj0vZKDMMIAEc,50085
 diffusers/callbacks.py,sha256=9NgQ7QkUncjVR8pbCHEd-PYvf5xKD-QRDmOoAjyzoXY,10282
 diffusers/configuration_utils.py,sha256=sf1KSxuGJSPo9beQvEXt8PolG0OVXdYF9WDRYssDGug,34942
 diffusers/dependency_versions_check.py,sha256=PcT_deWuvIKrNkjkCnQKi0ZTWCl77tHC02lhttbqQHM,1271
@@ -58,7 +58,7 @@ diffusers/models/__init__.py,sha256=EYlrNtK3-O_PuQqkxi1pJ2JmOPZ0eIpX6Q2HKBl_z-w,
 diffusers/models/activations.py,sha256=qxdn6OROfUvxyxgpm6M2VDKeJxH6mDsUI_xP4S3iw6s,6511
 diffusers/models/adapter.py,sha256=NDnqBqD53fg2fWWHt-LUHDyhuH6J-4R7PoStr2ggp-4,24507
 diffusers/models/attention.py,sha256=wxrklJ6tDweVrTFEJ2oCbcUIfdRha4irPQ602wNTEMo,73123
-diffusers/models/attention_dispatch.py,sha256=e7Bm5EWvvdPw0CT-NDvndHD1QahIOmGB9u2-g-rrahM,42265
+diffusers/models/attention_dispatch.py,sha256=Bk_-N04IIB2jN6y5gFo4yBNxPjOBtxpoYkOm1hlZFME,42798
 diffusers/models/attention_flax.py,sha256=NJTCmsALDnRScOy2EG7r6fZGXaNrJGBVTHQxAluWZEs,20392
 diffusers/models/attention_processor.py,sha256=1WrbKRIhcciXVorgI126bFQJuaQXbeYZbVY8b1bJ35A,239691
 diffusers/models/auto_model.py,sha256=6H0dnsr9atn-kgsEjnjLVpS6f3kzFDXCLhKUusxXdQE,11147
@@ -139,7 +139,7 @@ diffusers/models/transformers/transformer_flux.py,sha256=TLEdwpt-9u-sDUpuo14T91C
 diffusers/models/transformers/transformer_hidream_image.py,sha256=ciuyPRprws2K77apFzuLHmDZADOJ3AzAwsVze3FmR_E,39322
 diffusers/models/transformers/transformer_hunyuan_video.py,sha256=DMVzgTdQv0jrSiJA_qnt4eOcbRfDC_K1W4O9vvrWphw,47084
 diffusers/models/transformers/transformer_hunyuan_video_framepack.py,sha256=p9RgBqN5SWWqQU_wvEmaPZ2W5sCsYZoiGblqFKg6GpM,18698
-diffusers/models/transformers/transformer_ltx.py,sha256=7znlVN1UPdXBR91OXN_zyiwHt895XF71gCocV-BWpVc,22162
+diffusers/models/transformers/transformer_ltx.py,sha256=bFn8QbJG7O_VDCSIidQl-ovkl3jFEWQG4AT3m9By674,22200
 diffusers/models/transformers/transformer_lumina2.py,sha256=ACu9X7vatGMoiKSZKkEOcbEQjIvZwmqZ6nmHaSkD9Wo,22075
 diffusers/models/transformers/transformer_mochi.py,sha256=FbNpuQR3MrH7I7CU8tJ_8Nf00Q2VB5hNUpqwuYXS50Y,18521
 diffusers/models/transformers/transformer_omnigen.py,sha256=nxmuNqRXRG51Mw2G8BNu-DZrRKWMDc3G99DFgPS9yZA,20029
@@ -147,8 +147,8 @@ diffusers/models/transformers/transformer_qwenimage.py,sha256=Jzhx2Pf8odX8bRQLhP
 diffusers/models/transformers/transformer_sd3.py,sha256=2Aw1Di240iPngdlg_TZewlgi4WT2fGAOpB5XFa3RmO0,19232
 diffusers/models/transformers/transformer_skyreels_v2.py,sha256=PpHVausTXRFhm1gQJy_AFdzjtBb3NNThd2oqoPp4pAw,25733
 diffusers/models/transformers/transformer_temporal.py,sha256=GMn5WUbWWX7ZvyqVhO12g6bID7dnrMndYrDN-UZEI0Q,16812
-diffusers/models/transformers/transformer_wan.py,sha256=Y3e_j2CUZ0fce9yrkL7rYmDdAazigulkqeYEZEY1Mec,28852
-diffusers/models/transformers/transformer_wan_vace.py,sha256=KISLOLoYGRlPusyLOJEzjYdckFxFm_adv7b7Jg6jqYM,16477
+diffusers/models/transformers/transformer_wan.py,sha256=4dvIriHCdVVBf3H6n9vS5Ue9BArR92ZZpFTVYAQWrsA,28884
+diffusers/models/transformers/transformer_wan_vace.py,sha256=Bqaktjz0ETpwsta5jWvfP8fb1fsALRzW_sAGYcfY4qQ,16509
 diffusers/models/unets/__init__.py,sha256=srYFA7zEcDY7LxyUB2jz3TdRgsLz8elrWCpT6Y4YXuU,695
 diffusers/models/unets/unet_1d.py,sha256=tmSBsH3cPzNj9xCQt1zrV2lApufLEB2xiFTcXulE1Wo,10853
 diffusers/models/unets/unet_1d_blocks.py,sha256=LIuM8MwkcJ0n8wYwS6FBGYSMfl9wYv0-TbQ6FHO6A7k,26829
@@ -197,7 +197,7 @@ diffusers/pipelines/free_init_utils.py,sha256=SHrGV68cii9sYCKZLbIdBEOB5tANOVbN9v
 diffusers/pipelines/free_noise_utils.py,sha256=SlcvpUInyDDOOq7CkzXsjbcsC3Z7nY_JBzy6MJorHc4,29691
 diffusers/pipelines/onnx_utils.py,sha256=oTRc_iLHEKpf_IGFw_ka1bloAI-XUa6_ASMLW2LAH4w,8810
 diffusers/pipelines/pipeline_flax_utils.py,sha256=u8wFa0wdJ3CCCVpZdSLCAPDambSNGNLcyArEhiLwuAM,27026
-diffusers/pipelines/pipeline_loading_utils.py,sha256=RDbQqtx4F2Oog8b7jYC3vm-GfSZpa1UAUBQQVnfe_G4,48069
+diffusers/pipelines/pipeline_loading_utils.py,sha256=B7A8P0XoX4aCRBv0n6GdivzoDrX0m29nd7gF3GMXHaA,48489
 diffusers/pipelines/pipeline_utils.py,sha256=u-KRiPcVriYKHvEL_dT3PvmlBDCQBiqmFUVnexXx8oE,105794
 diffusers/pipelines/transformers_loading_utils.py,sha256=98wKUHN89Q1nmmat046hgQxLDlnZNj9Ww4TLB5W52pQ,5281
 diffusers/pipelines/allegro/__init__.py,sha256=T1MLZgDf8Fhh6YunF8a4Ta6NNIqneWsJIvmBhiy1ABM,1290
@@ -695,9 +695,9 @@ diffusers/utils/testing_utils.py,sha256=LSFTPVApvPOK9EHJYmgrUSRXzyYuHXFmEWHzvgF-
 diffusers/utils/torch_utils.py,sha256=ogvNvRfOryc5riP-Ph_nopGEsuZtmIszMIDPrkSgnAY,7775
 diffusers/utils/typing_utils.py,sha256=yeuCJmb1t5n5rG1JRPJo33KO7tg_m9ZwSXQcPKiKyFA,3400
 diffusers/utils/versions.py,sha256=-e7XW1TzZ-tsRo9PMQHp-hNGYHuVDFzLtwg3uAJzqdI,4333
-diffusers-0.35.1.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-diffusers-0.35.1.dist-info/METADATA,sha256=2YroDyHqhHv9IcdMEoodiSq5KijgbvS9mrwCfNOJFB8,20120
-diffusers-0.35.1.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-diffusers-0.35.1.dist-info/entry_points.txt,sha256=_1bvshKV_6_b63_FAkcUs9W6tUKGeIoQ3SHEZsovEWs,72
-diffusers-0.35.1.dist-info/top_level.txt,sha256=axJl2884vMSvhzrFrSoht36QXA_6gZN9cKtg4xOO72o,10
-diffusers-0.35.1.dist-info/RECORD,,
+diffusers-0.35.2.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+diffusers-0.35.2.dist-info/METADATA,sha256=RLXSlGdSKETBelaMUMUMPTXpjM3FaqRn3AiCqlBwvFc,20120
+diffusers-0.35.2.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+diffusers-0.35.2.dist-info/entry_points.txt,sha256=_1bvshKV_6_b63_FAkcUs9W6tUKGeIoQ3SHEZsovEWs,72
+diffusers-0.35.2.dist-info/top_level.txt,sha256=axJl2884vMSvhzrFrSoht36QXA_6gZN9cKtg4xOO72o,10
+diffusers-0.35.2.dist-info/RECORD,,