Add files using upload-large-folder tool

onnx/up_blocks.0/attentions.2.transformer_blocks.8.norm3.weight

@@ -0,0 +1,3 @@
+M:f:#:r::D:N::\:I:J:V:n:G:\:d:P:R:9:
+:N:z:t:n:�9V:V:D:Y:M:F:
+:�:Z:P:f::S:Y:+:?:_:J:0:6:V:\:F:t::U:*:�:P:�:Y:}:0:E:^:f:�9�:F:R:V:_:d:M:<:m:Q:O:<:a:^:8:�:�:>:8:L:(:M:Y:V:Z:R::L:�:Z:�:�:-:-:X::V:^:I:1:E:f:J:&:n:X:e:5:\:>:Q:f:f:u:�:\:Y:9:F:1:;:f:z:\:E:6:f:j:c:Q:l:,::0:^:�:K:�:\:o:z:\:.:�:�:a:O:':�:L:l:�:-:=:V::]:p:m:U:f:d:Q:Q:E:]:L:f:l:\:u:E:�:�9O:z:M:(:f:l:V:]:P:Z:^:(:M:>:p:>:f:N:O:l:t::f:.:m:X:�:P:P:x:�:(:S:N:N:I:�:*:[:�:d:w:P:E::(:M::X:S:u:n:F:\:�:y:(:z:�:M:�9I:^:l:X:F:u:f:{:M:F:u:\:6:u:P:i:>:u:u:V:U:@:\:�:d:G:e:6:�:>:^:^:/:E::f:z:\::u::X:F:V::t:^:M:M:\:f:�9d:n:<:V:Z:^:x:>:]:F:4:+:m:\:t:~:1:l:S:(:p:J:>:E:,:::K:T:u:n:<:=:�:u:N:<:V:Q:6:d:Z:r:i:N:j:f:K:l::R:9:�:6:z:(:U:E:E:N:M:<:1::m:^:N:n:�:.:(:�:j:�:�:I:I:f:d:M::Q:Q:�:S:Y:n::>:*:_:g:�9P:(:N:P:h:F:d:V:F:�:n:�:S:\:K:^:0:F:5:\:>:?:]:;:t:T:�:.:>::M:\:<:N:F::]:�:>:X:;:M:�:\:K:o:\:^:M:M:@:E:l:R:F::�:4:m:l:D:<:R:+:�:�:<:~:-:�:}:<:(:S:l:D:~:u:d:M:1:a:L:�9Z:2:K:f:F:0:R:+:a:z:K:6:<:d:\:`:r:<:%:N:-:P:N:-:}:<:z::f:N:l:�:p::6:L:r:K:K:F:�:D:V:>:u:>:d:�:2:>:c:F:4:f:L:E:S:K:T:>:F::V:O::>:l:X:d:.:>::G:+:o:6:^:V::M:�8>:K:Y:;:�9j:P:�:]:f:<:^:F:t:g:p:|:<:V:j:+:n:F::`:;:`:M:V:M:m:0:t::+:�:d::0:=:N:\:m:<:G:V:e:%:X:E:F:>:n:F:^:Y:a:0:�:`:�::�:<:+:z:m:U:Z:&:P:Z:P:L::<:^:<:P:0:�:L:�:\:O:\:v:>:

src/cache_diffusion/cachify.py

@@ -0,0 +1,144 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: MIT
+#
+# Permission is hereby granted, free of charge, to any person obtaining a
+# copy of this software and associated documentation files (the "Software"),
+# to deal in the Software without restriction, including without limitation
+# the rights to use, copy, modify, merge, publish, distribute, sublicense,
+# and/or sell copies of the Software, and to permit persons to whom the
+# Software is furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+# DEALINGS IN THE SOFTWARE.
+
+import fnmatch
+from contextlib import contextmanager
+
+from diffusers.models.attention import BasicTransformerBlock, JointTransformerBlock
+from diffusers.models.transformers.pixart_transformer_2d import PixArtTransformer2DModel
+from diffusers.models.transformers.transformer_sd3 import SD3Transformer2DModel
+from diffusers.models.unets.unet_2d_blocks import (
+    CrossAttnDownBlock2D,
+    CrossAttnUpBlock2D,
+    DownBlock2D,
+    UNetMidBlock2DCrossAttn,
+    UpBlock2D,
+)
+from diffusers.models.unets.unet_2d_condition import UNet2DConditionModel
+from diffusers.models.unets.unet_3d_blocks import (
+    CrossAttnDownBlockSpatioTemporal,
+    CrossAttnUpBlockSpatioTemporal,
+    DownBlockSpatioTemporal,
+    UNetMidBlockSpatioTemporal,
+    UpBlockSpatioTemporal,
+)
+from diffusers.models.unets.unet_spatio_temporal_condition import UNetSpatioTemporalConditionModel
+
+from .module import CachedModule
+from .utils import replace_module
+
+CACHED_PIPE = {
+    UNet2DConditionModel: (
+        DownBlock2D,
+        CrossAttnDownBlock2D,
+        UNetMidBlock2DCrossAttn,
+        CrossAttnUpBlock2D,
+        UpBlock2D,
+    ),
+    PixArtTransformer2DModel: (BasicTransformerBlock),
+    UNetSpatioTemporalConditionModel: (
+        CrossAttnDownBlockSpatioTemporal,
+        DownBlockSpatioTemporal,
+        UpBlockSpatioTemporal,
+        CrossAttnUpBlockSpatioTemporal,
+        UNetMidBlockSpatioTemporal,
+    ),
+    SD3Transformer2DModel: (JointTransformerBlock),
+}
+
+
+def _apply_to_modules(model, action, modules=None, config_list=None):
+    if hasattr(model, "use_trt_infer") and model.use_trt_infer:
+        for key, module in model.engines.items():
+            if isinstance(module, CachedModule):
+                action(module)
+            elif config_list:
+                for config in config_list:
+                    if _pass(key, config["wildcard_or_filter_func"]):
+                        model.engines[key] = CachedModule(module, config["select_cache_step_func"])
+    else:
+        for name, module in model.named_modules():
+            if isinstance(module, CachedModule):
+                action(module)
+            elif modules and config_list:
+                for config in config_list:
+                    if _pass(name, config["wildcard_or_filter_func"]) and isinstance(
+                        module, modules
+                    ):
+                        replace_module(
+                            model,
+                            name,
+                            CachedModule(module, config["select_cache_step_func"]),
+                        )
+
+
+def cachify(model, config_list, modules):
+    def cache_action(module):
+        pass  # No action needed, caching is handled in the loop itself
+
+    _apply_to_modules(model, cache_action, modules, config_list)
+
+
+def disable(pipe):
+    model = get_model(pipe)
+    _apply_to_modules(model, lambda module: module.disable_cache())
+
+
+def enable(pipe):
+    model = get_model(pipe)
+    _apply_to_modules(model, lambda module: module.enable_cache())
+
+
+def reset_status(pipe):
+    model = get_model(pipe)
+    _apply_to_modules(model, lambda module: setattr(module, "cur_step", 0))
+
+
+def _pass(name, wildcard_or_filter_func):
+    if isinstance(wildcard_or_filter_func, str):
+        return fnmatch.fnmatch(name, wildcard_or_filter_func)
+    elif callable(wildcard_or_filter_func):
+        return wildcard_or_filter_func(name)
+    else:
+        raise NotImplementedError(f"Unsupported type {type(wildcard_or_filter_func)}")
+
+
+def get_model(pipe):
+    if hasattr(pipe, "unet"):
+        return pipe.unet
+    elif hasattr(pipe, "transformer"):
+        return pipe.transformer
+    else:
+        raise KeyError
+
+
+@contextmanager
+def infer(pipe):
+    try:
+        yield pipe
+    finally:
+        reset_status(pipe)
+
+
+def prepare(pipe, config_list):
+    model = get_model(pipe)
+    assert model.__class__ in CACHED_PIPE.keys(), f"{model.__class__} is not supported!"
+    cachify(model, config_list, CACHED_PIPE[model.__class__])