cache-dit 1.0.2__py3-none-any.whl → 1.0.3__py3-none-any.whl

cache_dit/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '1.0.2'
-__version_tuple__ = version_tuple = (1, 0, 2)
+__version__ = version = '1.0.3'
+__version_tuple__ = version_tuple = (1, 0, 3)
 
 __commit_id__ = commit_id = None

cache_dit/cache_factory/cache_blocks/pattern_3_4_5.py

@@ -33,14 +33,14 @@ class CachedBlocks_Pattern_3_4_5(CachedBlocks_Pattern_Base):
                 *args,
                 **kwargs,
             )
-            hidden_states, new_encoder_hidden_states = self._process_outputs(
-                hidden_states
+            hidden_states, new_encoder_hidden_states = (
+                self._process_block_outputs(hidden_states)
             )
 
         return hidden_states, new_encoder_hidden_states
 
     @torch.compiler.disable
-    def _process_outputs(
+    def _process_block_outputs(
         self, hidden_states: torch.Tensor | tuple
     ) -> tuple[torch.Tensor, torch.Tensor | None]:
         # Process the outputs for the block.
@@ -66,7 +66,7 @@ class CachedBlocks_Pattern_3_4_5(CachedBlocks_Pattern_Base):
         return hidden_states, new_encoder_hidden_states
 
     @torch.compiler.disable
-    def _forward_outputs(
+    def _process_forward_outputs(
         self,
         hidden_states: torch.Tensor,
         new_encoder_hidden_states: torch.Tensor | None,
@@ -100,7 +100,7 @@ class CachedBlocks_Pattern_3_4_5(CachedBlocks_Pattern_Base):
                 *args,
                 **kwargs,
             )
-            return self._forward_outputs(
+            return self._process_forward_outputs(
                 hidden_states, new_encoder_hidden_states
             )
 
@@ -227,7 +227,10 @@ class CachedBlocks_Pattern_3_4_5(CachedBlocks_Pattern_Base):
 
         torch._dynamo.graph_break()
 
-        return self._forward_outputs(hidden_states, new_encoder_hidden_states)
+        return self._process_forward_outputs(
+            hidden_states,
+            new_encoder_hidden_states,
+        )
 
     def call_Fn_blocks(
         self,
@@ -242,8 +245,8 @@ class CachedBlocks_Pattern_3_4_5(CachedBlocks_Pattern_Base):
                 *args,
                 **kwargs,
             )
-            hidden_states, new_encoder_hidden_states = self._process_outputs(
-                hidden_states
+            hidden_states, new_encoder_hidden_states = (
+                self._process_block_outputs(hidden_states)
             )
 
         return hidden_states, new_encoder_hidden_states
@@ -263,8 +266,8 @@ class CachedBlocks_Pattern_3_4_5(CachedBlocks_Pattern_Base):
                 **kwargs,
             )
 
-            hidden_states, new_encoder_hidden_states = self._process_outputs(
-                hidden_states
+            hidden_states, new_encoder_hidden_states = (
+                self._process_block_outputs(hidden_states)
             )
 
         # compute hidden_states residual
@@ -296,8 +299,8 @@ class CachedBlocks_Pattern_3_4_5(CachedBlocks_Pattern_Base):
                 **kwargs,
             )
 
-            hidden_states, new_encoder_hidden_states = self._process_outputs(
-                hidden_states
+            hidden_states, new_encoder_hidden_states = (
+                self._process_block_outputs(hidden_states)
             )
 
         return hidden_states, new_encoder_hidden_states

cache_dit/cache_factory/cache_blocks/pattern_base.py

@@ -135,7 +135,7 @@ class CachedBlocks_Pattern_Base(torch.nn.Module):
         return hidden_states, encoder_hidden_states
 
     @torch.compiler.disable
-    def _process_outputs(
+    def _process_block_outputs(
         self,
         hidden_states: torch.Tensor | tuple,
         encoder_hidden_states: torch.Tensor | None,
@@ -150,7 +150,7 @@ class CachedBlocks_Pattern_Base(torch.nn.Module):
         return hidden_states, encoder_hidden_states
 
     @torch.compiler.disable
-    def _forward_outputs(
+    def _process_forward_outputs(
         self,
         hidden_states: torch.Tensor,
         encoder_hidden_states: torch.Tensor | None,
@@ -185,7 +185,10 @@ class CachedBlocks_Pattern_Base(torch.nn.Module):
                 *args,
                 **kwargs,
             )
-            return self._forward_outputs(hidden_states, encoder_hidden_states)
+            return self._process_forward_outputs(
+                hidden_states,
+                encoder_hidden_states,
+            )
 
         original_hidden_states = hidden_states
         # Call first `n` blocks to process the hidden states for
@@ -304,7 +307,10 @@ class CachedBlocks_Pattern_Base(torch.nn.Module):
         # patch cached stats for blocks or remove it.
         torch._dynamo.graph_break()
 
-        return self._forward_outputs(hidden_states, encoder_hidden_states)
+        return self._process_forward_outputs(
+            hidden_states,
+            encoder_hidden_states,
+        )
 
     @torch.compiler.disable
     def _is_parallelized(self):
@@ -379,7 +385,7 @@ class CachedBlocks_Pattern_Base(torch.nn.Module):
                 *args,
                 **kwargs,
             )
-            hidden_states, encoder_hidden_states = self._process_outputs(
+            hidden_states, encoder_hidden_states = self._process_block_outputs(
                 hidden_states, encoder_hidden_states
             )
 
@@ -401,7 +407,7 @@ class CachedBlocks_Pattern_Base(torch.nn.Module):
                 *args,
                 **kwargs,
             )
-            hidden_states, encoder_hidden_states = self._process_outputs(
+            hidden_states, encoder_hidden_states = self._process_block_outputs(
                 hidden_states, encoder_hidden_states
             )
 
@@ -445,7 +451,7 @@ class CachedBlocks_Pattern_Base(torch.nn.Module):
                 *args,
                 **kwargs,
             )
-            hidden_states, encoder_hidden_states = self._process_outputs(
+            hidden_states, encoder_hidden_states = self._process_block_outputs(
                 hidden_states, encoder_hidden_states
             )
 

cache_dit/cache_factory/cache_contexts/cache_context.py

@@ -38,6 +38,10 @@ class BasicCacheConfig:
     #     DBCache does not apply the caching strategy when the number of running steps is less than
     #     or equal to this value, ensuring the model sufficiently learns basic features during warmup.
     max_warmup_steps: int = 8  # DON'T Cache in warmup steps
+    # warmup_interval (`int`, *required*, defaults to 1):
+    #     Skip interval in warmup steps, e.g., when warmup_interval is 2, only 0, 2, 4, ... steps
+    #     in warmup steps will be computed, others will use dynamic cache.
+    warmup_interval: int = 1  # skip interval in warmup steps
     # max_cached_steps (`int`, *required*, defaults to -1):
     #     DBCache disables the caching strategy when the previous cached steps exceed this value to
     #     prevent precision degradation.
@@ -71,6 +75,7 @@ class BasicCacheConfig:
             f"DBCACHE_F{self.Fn_compute_blocks}"
             f"B{self.Bn_compute_blocks}_"
             f"W{self.max_warmup_steps}"
+            f"I{self.warmup_interval}"
             f"M{max(0, self.max_cached_steps)}"
             f"MC{max(0, self.max_continuous_cached_steps)}_"
             f"R{self.residual_diff_threshold}"
@@ -346,5 +351,15 @@ class CachedContext:
         # CFG steps: 1, 3, 5, 7, ...
         return self.get_current_transformer_step() % 2 != 0
 
+    @property
+    def warmup_steps(self) -> List[int]:
+        return list(
+            range(
+                0,
+                self.cache_config.max_warmup_steps,
+                self.cache_config.warmup_interval,
+            )
+        )
+
     def is_in_warmup(self):
-        return self.get_current_step() < self.cache_config.max_warmup_steps
+        return self.get_current_step() in self.warmup_steps

cache_dit/cache_factory/cache_interface.py

@@ -86,6 +86,9 @@ def enable_cache(
                 max_warmup_steps (`int`, *required*, defaults to 8):
                     DBCache does not apply the caching strategy when the number of running steps is less than
                     or equal to this value, ensuring the model sufficiently learns basic features during warmup.
+                warmup_interval (`int`, *required*, defaults to 1):
+                    Skip interval in warmup steps, e.g., when warmup_interval is 2, only 0, 2, 4, ... steps
+                    in warmup steps will be computed, others will use dynamic cache.
                 max_cached_steps (`int`, *required*, defaults to -1):
                     DBCache disables the caching strategy when the previous cached steps exceed this value to
                     prevent precision degradation.

cache_dit/cache_factory/utils.py

@@ -1,14 +1,12 @@
 import yaml
-from cache_dit.cache_factory import CacheType
 
 
 def load_cache_options_from_yaml(yaml_file_path):
     try:
         with open(yaml_file_path, "r") as f:
-            config = yaml.safe_load(f)
+            kwargs: dict = yaml.safe_load(f)
 
         required_keys = [
-            "cache_type",
             "max_warmup_steps",
             "max_cached_steps",
             "Fn_compute_blocks",
@@ -16,34 +14,36 @@ def load_cache_options_from_yaml(yaml_file_path):
             "residual_diff_threshold",
         ]
         for key in required_keys:
-            if key not in config:
+            if key not in kwargs:
                 raise ValueError(
                     f"Configuration file missing required item: {key}"
                 )
 
-        # Convert cache_type to CacheType enum
-        if isinstance(config["cache_type"], str):
-            try:
-                config["cache_type"] = CacheType[config["cache_type"]]
-            except KeyError:
-                valid_types = [ct.name for ct in CacheType]
-                raise ValueError(
-                    f"Invalid cache_type value: {config['cache_type']}, "
-                    f"valid values are: {valid_types}"
+        cache_context_kwargs = {}
+        if kwargs.get("enable_taylorseer", False):
+            from cache_dit.cache_factory.cache_contexts.calibrators import (
+                TaylorSeerCalibratorConfig,
+            )
+
+            cache_context_kwargs["calibrator_config"] = (
+                TaylorSeerCalibratorConfig(
+                    enable_calibrator=kwargs.pop("enable_taylorseer"),
+                    enable_encoder_calibrator=kwargs.pop(
+                        "enable_encoder_taylorseer", False
+                    ),
+                    calibrator_cache_type=kwargs.pop(
+                        "taylorseer_cache_type", "residual"
+                    ),
+                    taylorseer_order=kwargs.pop("taylorseer_order", 1),
                 )
-        elif not isinstance(config["cache_type"], CacheType):
-            raise ValueError(
-                f"cache_type must be a string or CacheType enum, "
-                f"got: {type(config['cache_type'])}"
             )
 
-        # Handle default value for taylorseer_kwargs
-        if "taylorseer_kwargs" not in config and config.get(
-            "enable_taylorseer", False
-        ):
-            config["taylorseer_kwargs"] = {"n_derivatives": 2}
+        from cache_dit.cache_factory.cache_contexts import BasicCacheConfig
+
+        cache_context_kwargs["cache_config"] = BasicCacheConfig()
+        cache_context_kwargs["cache_config"].update(**kwargs)
 
-        return config
+        return cache_context_kwargs
 
     except FileNotFoundError:
         raise FileNotFoundError(

{cache_dit-1.0.2.dist-info → cache_dit-1.0.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cache_dit
-Version: 1.0.2
+Version: 1.0.3
 Summary: A Unified, Flexible and Training-free Cache Acceleration Framework for 🤗Diffusers.
 Author: DefTruth, vipshop.com, etc.
 Maintainer: DefTruth, vipshop.com, etc
@@ -247,6 +247,8 @@ For more advanced features such as **Unified Cache APIs**, **Forward Pattern Mat
 - [⚙️Torch Compile](./docs/User_Guide.md#️torch-compile)
 - [📚API Documents](./docs/User_Guide.md#api-documentation)
 
+
+
 ## 👋Contribute 
 <div id="contribute"></div>
 
@@ -260,8 +262,13 @@ How to contribute? Star ⭐️ this repo to support us or check [CONTRIBUTE.md](
     <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=vipshop/cache-dit&type=Date" width=400px />
   </picture>
 </a>
+
 </div>
 
+## 🎉Projects Using CacheDiT
+
+Here is a curated list of open-source projects integrating **CacheDiT**, including popular repositories like [jetson-containers](https://github.com/dusty-nv/jetson-containers/blob/master/packages/diffusion/cache_edit/build.sh) ![](https://img.shields.io/github/stars/dusty-nv/jetson-containers.svg), [flux-fast](https://github.com/huggingface/flux-fast) ![](https://img.shields.io/github/stars/huggingface/flux-fast.svg), and [sdnext](https://github.com/vladmandic/sdnext/blob/dev/modules/cachedit.py) ![](https://img.shields.io/github/stars/vladmandic/sdnext.svg). **CacheDiT** has also been **recommended** by [Wan2.2](https://github.com/Wan-Video/Wan2.2) ![](https://img.shields.io/github/stars/Wan-Video/Wan2.2.svg), [Qwen-Image-Lightning](https://github.com/ModelTC/Qwen-Image-Lightning) ![](https://img.shields.io/github/stars/ModelTC/Qwen-Image-Lightning.svg), [Qwen-Image](https://github.com/QwenLM/Qwen-Image) ![](https://img.shields.io/github/stars/QwenLM/Qwen-Image.svg), and <a href="https://huggingface.co/docs/diffusers/main/en/optimization/cache_dit"><img src="https://img.shields.io/badge/🤗Diffusers-ecosystem-yellow.svg"></a> ![](https://img.shields.io/github/stars/huggingface/diffusers.svg), among others. We would be grateful if you could let us know if you have used CacheDiT.
+
 ## ©️Acknowledgements
 
 <div id="Acknowledgements"></div>

{cache_dit-1.0.2.dist-info → cache_dit-1.0.3.dist-info}/RECORD

@@ -1,14 +1,14 @@
 cache_dit/__init__.py,sha256=sHRg0swXZZiw6lvSQ53fcVtN9JRayx0az2lXAz5OOGI,1510
-cache_dit/_version.py,sha256=ZTgKq8LPNy3l9uR2ke-VtLhvvl5l71frQ9wO76n1L5k,704
+cache_dit/_version.py,sha256=l8k828IdTfzXAlmx4oT8GsiIf2eeMAlFDALjoYk-jrU,704
 cache_dit/logger.py,sha256=0zsu42hN-3-rgGC_C29ms1IvVpV4_b4_SwJCKSenxBE,4304
 cache_dit/utils.py,sha256=AyYRwi5XBxYBH4GaXxOxv9-X24Te_IYOYwh54t_1d3A,10674
 cache_dit/cache_factory/.gitignore,sha256=5Cb-qT9wsTUoMJ7vACDF7ZcLpAXhi5v-xdcWSRit988,23
 cache_dit/cache_factory/__init__.py,sha256=vy9I6Ofkj9jWeUoOvh-cY5a9QlDDKfj2FVPlVTf7BeA,1390
-cache_dit/cache_factory/cache_interface.py,sha256=KseSPyZ9D3m6pmpE7k-uYr0wfBI-hhscG1Nw54GCHxk,12316
+cache_dit/cache_factory/cache_interface.py,sha256=fJgsOSR_lP0cvNDrR0zMLLoZBZC6tLAQaPQs_oo2R1o,12577
 cache_dit/cache_factory/cache_types.py,sha256=ooukxQRG55uTLmaZ0SKw6gIeY6SQHhMxkbv55uj2Sqk,991
 cache_dit/cache_factory/forward_pattern.py,sha256=FumlCuZ-TSmSYH0hGBHctSJ-oGLCftdZjLygqhsmdR4,2258
 cache_dit/cache_factory/params_modifier.py,sha256=zYJJsInTYCaYHBZ7mZJOP-PZnkSg3iN1WPewNOayXos,3628
-cache_dit/cache_factory/utils.py,sha256=XkVM9AXcB9zYq8-S8QKAsGz80r3tA6U3lBNGDGeHOe4,1871
+cache_dit/cache_factory/utils.py,sha256=mm8JNu6XG_w6nMYvv53TmugSb-l3W7l3Y4rJ2xBgktY,1891
 cache_dit/cache_factory/block_adapters/__init__.py,sha256=vM3aDMzPY79Tw4L0hlV2PdA3MFYomnf0eo0BGBo9P78,18087
 cache_dit/cache_factory/block_adapters/block_adapters.py,sha256=2TVK_KqiYXC7AKZ2s07fzdOzUoeUBc9P1SzQtLVzhf4,22249
 cache_dit/cache_factory/block_adapters/block_registers.py,sha256=2L7QeM4ygnaKQpC9PoJod0QRYyxidUKU2AYpysDCUwE,2572
@@ -17,11 +17,11 @@ cache_dit/cache_factory/cache_adapters/cache_adapter.py,sha256=HTyZdspd34G6QiJ2q
 cache_dit/cache_factory/cache_blocks/__init__.py,sha256=mivvm8YOfqT7YHs8y_MzGOGztPw8LxAqKGXuSRXxCv0,3032
 cache_dit/cache_factory/cache_blocks/offload_utils.py,sha256=wusgcqaCrwEjvv7Guy-6VXhNOgPPUrBV2sSVuRmGuvo,3513
 cache_dit/cache_factory/cache_blocks/pattern_0_1_2.py,sha256=ElMps6_7uI74tSF9GDR_dEI0bZEhdzcepM29xFWnYo8,428
-cache_dit/cache_factory/cache_blocks/pattern_3_4_5.py,sha256=rfq5-WEt-ErY28vcB4ur9E-uCb6BKP0S8v5lTw61ROk,10555
-cache_dit/cache_factory/cache_blocks/pattern_base.py,sha256=StNW2PyDiXEIxZd30byPUrZZ8jgSiuC_yrly2w7X2LQ,16176
+cache_dit/cache_factory/cache_blocks/pattern_3_4_5.py,sha256=mzs1S2YFwNAPMMTisTKbU6GA5m60J_20CAVy9OIWoMQ,10652
+cache_dit/cache_factory/cache_blocks/pattern_base.py,sha256=UeBYEz3hamO3CyVMj1KI7GnxRVQGBjQ5EJi90obVZyI,16306
 cache_dit/cache_factory/cache_blocks/pattern_utils.py,sha256=dGOC1tMMOvcbvEgx44eTESKn_jsv-0RZ3tRHPa3wmQ4,1315
 cache_dit/cache_factory/cache_contexts/__init__.py,sha256=N3SxFnluXk5q09nhSqKIJCVzEGWzySJWm-vic6dH79E,412
-cache_dit/cache_factory/cache_contexts/cache_context.py,sha256=3EhaMCz3VUQ_NF81VgYwWoSEGIvhScPxPYhjL1OcgxE,15240
+cache_dit/cache_factory/cache_contexts/cache_context.py,sha256=FXvrR3XZr4iIsKSTBngzaRM6_WxiHkRNQ3wAJz40kbk,15798
 cache_dit/cache_factory/cache_contexts/cache_manager.py,sha256=X99XnmiY-Us8D2pqJGPKxWcXAhQQpk3xdEWOOOYXIZ4,30465
 cache_dit/cache_factory/cache_contexts/calibrators/__init__.py,sha256=mzYXO8tbytGpJJ9rpPu20kMoj1Iu_7Ym9tjfzV8rA98,5574
 cache_dit/cache_factory/cache_contexts/calibrators/base.py,sha256=mn6ZBkChGpGwN5csrHTUGMoX6BBPvqHXSLbIExiW-EU,748
@@ -50,9 +50,9 @@ cache_dit/metrics/metrics.py,sha256=AZbQyoavE-djvyRUZ_EfCIrWSQbiWQFo7n2dhn7XptE,
 cache_dit/quantize/__init__.py,sha256=kWYoMAyZgBXu9BJlZjTQ0dRffW9GqeeY9_iTkXrb70A,59
 cache_dit/quantize/quantize_ao.py,sha256=Pr3u3Qr6qLvFkd8k-_rfcz4Mkjlg36U9BHG2t6Bl-6M,6301
 cache_dit/quantize/quantize_interface.py,sha256=2s_R7xPSKuJeFpEGeLwRxnq_CqJcBG3a3lzyW5wh-UM,1241
-cache_dit-1.0.2.dist-info/licenses/LICENSE,sha256=Dqb07Ik2dV41s9nIdMUbiRWEfDqo7-dQeRiY7kPO8PE,3769
-cache_dit-1.0.2.dist-info/METADATA,sha256=E6MkP_T9cwJEbqWE1DIRVkQLI7wLWr5zryY2poWgkyw,26766
-cache_dit-1.0.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-cache_dit-1.0.2.dist-info/entry_points.txt,sha256=FX2gysXaZx6NeK1iCLMcIdP8Q4_qikkIHtEmi3oWn8o,65
-cache_dit-1.0.2.dist-info/top_level.txt,sha256=ZJDydonLEhujzz0FOkVbO-BqfzO9d_VqRHmZU-3MOZo,10
-cache_dit-1.0.2.dist-info/RECORD,,
+cache_dit-1.0.3.dist-info/licenses/LICENSE,sha256=Dqb07Ik2dV41s9nIdMUbiRWEfDqo7-dQeRiY7kPO8PE,3769
+cache_dit-1.0.3.dist-info/METADATA,sha256=gPY4pnvl4dvTTu7Twv6unzEesu1fXCDlGNMlSdFP3Lc,28103
+cache_dit-1.0.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+cache_dit-1.0.3.dist-info/entry_points.txt,sha256=FX2gysXaZx6NeK1iCLMcIdP8Q4_qikkIHtEmi3oWn8o,65
+cache_dit-1.0.3.dist-info/top_level.txt,sha256=ZJDydonLEhujzz0FOkVbO-BqfzO9d_VqRHmZU-3MOZo,10
+cache_dit-1.0.3.dist-info/RECORD,,