cache-dit 0.2.27__py3-none-any.whl → 0.2.29__py3-none-any.whl

cache_dit/cache_factory/cache_interface.py

@@ -1,9 +1,11 @@
+import torch
 from typing import Any, Tuple, List
 from diffusers import DiffusionPipeline
 from cache_dit.cache_factory.cache_types import CacheType
 from cache_dit.cache_factory.block_adapters import BlockAdapter
 from cache_dit.cache_factory.block_adapters import BlockAdapterRegistry
 from cache_dit.cache_factory.cache_adapters import CachedAdapter
+from cache_dit.cache_factory.cache_contexts import CachedContextManager
 
 from cache_dit.logger import init_logger
 
@@ -12,7 +14,7 @@ logger = init_logger(__name__)
 
 def enable_cache(
     # DiffusionPipeline or BlockAdapter
-    pipe_or_adapter: DiffusionPipeline | BlockAdapter | Any,
+    pipe_or_adapter: DiffusionPipeline | BlockAdapter,
     # Cache context kwargs
     Fn_compute_blocks: int = 8,
     Bn_compute_blocks: int = 0,
@@ -21,7 +23,7 @@ def enable_cache(
     max_continuous_cached_steps: int = -1,
     residual_diff_threshold: float = 0.08,
     # Cache CFG or not
-    do_separate_cfg: bool = False,
+    enable_spearate_cfg: bool = False,
     cfg_compute_first: bool = False,
     cfg_diff_compute_separate: bool = True,
     # Hybird TaylorSeer
@@ -30,7 +32,7 @@ def enable_cache(
     taylorseer_cache_type: str = "residual",
     taylorseer_order: int = 2,
     **other_cache_context_kwargs,
-) -> DiffusionPipeline | Any:
+) -> BlockAdapter:
     r"""
     Unified Cache API for  almost Any Diffusion Transformers (with Transformer Blocks
     that match the specific Input and Output patterns).
@@ -64,9 +66,9 @@ def enable_cache(
         residual_diff_threshold (`float`, *required*, defaults to 0.08):
             he value of residual diff threshold, a higher value leads to faster performance at the
             cost of lower precision.
-        do_separate_cfg (`bool`, *required*,  defaults to False):
+        enable_spearate_cfg (`bool`, *required*,  defaults to False):
             Whether to do separate cfg or not, such as Wan 2.1, Qwen-Image. For model that fused CFG
-            and non-CFG into single forward step, should set do_separate_cfg as False, for example:
+            and non-CFG into single forward step, should set enable_spearate_cfg as False, for example:
             CogVideoX, HunyuanVideo, Mochi, etc.
         cfg_compute_first (`bool`, *required*,  defaults to False):
             Compute cfg forward first or not, default False, namely, 0, 2, 4, ..., -> non-CFG step;
@@ -89,7 +91,7 @@ def enable_cache(
             The order of taylorseer, higher values of n_derivatives will lead to longer computation time,
             but may improve precision significantly.
         other_cache_kwargs: (`dict`, *optional*, defaults to {})
-            Other cache context kwargs, please check https://github.com/vipshop/cache-dit/blob/main/src/cache_dit/cache_factory/cache_context.py
+            Other cache context kwargs, please check https://github.com/vipshop/cache-dit/blob/main/src/cache_dit/cache_factory/cache_contexts/cache_context.py
             for more details.
 
     Examples:
@@ -104,6 +106,10 @@ def enable_cache(
 
     # Collect cache context kwargs
     cache_context_kwargs = other_cache_context_kwargs.copy()
+    if cache_type := cache_context_kwargs.get("cache_type", None):
+        if cache_type == CacheType.NONE:
+            return pipe_or_adapter
+
     cache_context_kwargs["cache_type"] = CacheType.DBCache
     cache_context_kwargs["Fn_compute_blocks"] = Fn_compute_blocks
     cache_context_kwargs["Bn_compute_blocks"] = Bn_compute_blocks
@@ -113,7 +119,7 @@ def enable_cache(
         max_continuous_cached_steps
     )
     cache_context_kwargs["residual_diff_threshold"] = residual_diff_threshold
-    cache_context_kwargs["do_separate_cfg"] = do_separate_cfg
+    cache_context_kwargs["enable_spearate_cfg"] = enable_spearate_cfg
     cache_context_kwargs["cfg_compute_first"] = cfg_compute_first
     cache_context_kwargs["cfg_diff_compute_separate"] = (
         cfg_diff_compute_separate
@@ -123,25 +129,11 @@ def enable_cache(
         enable_encoder_taylorseer
     )
     cache_context_kwargs["taylorseer_cache_type"] = taylorseer_cache_type
-    if "taylorseer_kwargs" in cache_context_kwargs:
-        cache_context_kwargs["taylorseer_kwargs"][
-            "n_derivatives"
-        ] = taylorseer_order
-    else:
-        cache_context_kwargs["taylorseer_kwargs"] = {
-            "n_derivatives": taylorseer_order
-        }
+    cache_context_kwargs["taylorseer_order"] = taylorseer_order
 
-    if isinstance(pipe_or_adapter, BlockAdapter):
-        return CachedAdapter.apply(
-            pipe=None,
-            block_adapter=pipe_or_adapter,
-            **cache_context_kwargs,
-        )
-    elif isinstance(pipe_or_adapter, DiffusionPipeline):
+    if isinstance(pipe_or_adapter, (DiffusionPipeline, BlockAdapter)):
         return CachedAdapter.apply(
-            pipe=pipe_or_adapter,
-            block_adapter=None,
+            pipe_or_adapter,
             **cache_context_kwargs,
         )
     else:
@@ -152,6 +144,81 @@ def enable_cache(
         )
 
 
+def disable_cache(
+    # DiffusionPipeline or BlockAdapter
+    pipe_or_adapter: DiffusionPipeline | BlockAdapter,
+):
+    from cache_dit.cache_factory.cache_blocks.utils import (
+        remove_cached_stats,
+    )
+
+    def _disable_blocks(blocks: torch.nn.ModuleList):
+        if blocks is None:
+            return
+        if hasattr(blocks, "_forward_pattern"):
+            del blocks._forward_pattern
+        if hasattr(blocks, "_cache_context_kwargs"):
+            del blocks._cache_context_kwargs
+        remove_cached_stats(blocks)
+
+    def _disable_transformer(transformer: torch.nn.Module):
+        if transformer is None or not BlockAdapter.is_cached(transformer):
+            return
+        if original_forward := getattr(transformer, "_original_forward"):
+            transformer.forward = original_forward.__get__(transformer)
+            del transformer._original_forward
+        if hasattr(transformer, "_is_cached"):
+            del transformer._is_cached
+        if hasattr(transformer, "_forward_pattern"):
+            del transformer._forward_pattern
+        if hasattr(transformer, "_has_separate_cfg"):
+            del transformer._has_separate_cfg
+        if hasattr(transformer, "_cache_context_kwargs"):
+            del transformer._cache_context_kwargs
+        remove_cached_stats(transformer)
+        for blocks in BlockAdapter.find_blocks(transformer):
+            _disable_blocks(blocks)
+
+    def _disable_pipe(pipe: DiffusionPipeline):
+        if pipe is None or not BlockAdapter.is_cached(pipe):
+            return
+        if original_call := getattr(pipe, "_original_call"):
+            pipe.__class__.__call__ = original_call
+            del pipe.__class__._original_call
+        if cache_manager := getattr(pipe, "_cache_manager"):
+            assert isinstance(cache_manager, CachedContextManager)
+            cache_manager.clear_contexts()
+            del pipe._cache_manager
+        if hasattr(pipe, "_is_cached"):
+            del pipe.__class__._is_cached
+        if hasattr(pipe, "_cache_context_kwargs"):
+            del pipe._cache_context_kwargs
+        remove_cached_stats(pipe)
+
+    if isinstance(pipe_or_adapter, DiffusionPipeline):
+        pipe = pipe_or_adapter
+        _disable_pipe(pipe)
+        if hasattr(pipe, "transformer"):
+            _disable_transformer(pipe.transformer)
+        if hasattr(pipe, "transformer_2"):  # Wan 2.2
+            _disable_transformer(pipe.transformer_2)
+        pipe_cls_name = pipe.__class__.__name__
+        logger.warning(f"Cache Acceleration is disabled for: {pipe_cls_name}")
+    elif isinstance(pipe_or_adapter, BlockAdapter):
+        # BlockAdapter
+        adapter = pipe_or_adapter
+        BlockAdapter.assert_normalized(adapter)
+        _disable_pipe(adapter.pipe)
+        for transformer in BlockAdapter.flatten(adapter.transformer):
+            _disable_transformer(transformer)
+        for blocks in BlockAdapter.flatten(adapter.blocks):
+            _disable_blocks(blocks)
+        pipe_cls_name = adapter.pipe.__class__.__name__
+        logger.warning(f"Cache Acceleration is disabled for: {pipe_cls_name}")
+    else:
+        pass  # do nothing
+
+
 def supported_pipelines(
     **kwargs,
 ) -> Tuple[int, List[str]]:

cache_dit/cache_factory/patch_functors/functor_chroma.py

@@ -30,7 +30,7 @@ class ChromaPatchFunctor(PatchFunctor):
         blocks: torch.nn.ModuleList = None,
         **kwargs,
     ) -> ChromaTransformer2DModel:
-        if getattr(transformer, "_is_patched", False):
+        if hasattr(transformer, "_is_patched"):
             return transformer
 
         if blocks is None:

cache_dit/cache_factory/patch_functors/functor_flux.py

@@ -31,7 +31,7 @@ class FluxPatchFunctor(PatchFunctor):
         **kwargs,
     ) -> FluxTransformer2DModel:
 
-        if getattr(transformer, "_is_patched", False):
+        if hasattr(transformer, "_is_patched"):
             return transformer
 
         if blocks is None:

cache_dit/utils.py

@@ -5,9 +5,9 @@ import numpy as np
 from pprint import pprint
 from diffusers import DiffusionPipeline
 
-from typing import Dict, Any
+from typing import Dict, Any, List, Union
+from cache_dit.cache_factory import BlockAdapter
 from cache_dit.logger import init_logger
-from cache_dit.cache_factory import CacheType
 
 
 logger = init_logger(__name__)
@@ -30,9 +30,168 @@ class CacheStats:
 
 
 def summary(
-    pipe_or_module: DiffusionPipeline | torch.nn.Module | Any,
+    adapter_or_others: Union[
+        BlockAdapter,
+        DiffusionPipeline,
+        torch.nn.Module,
+    ],
     details: bool = False,
     logging: bool = True,
+    **kwargs,
+) -> List[CacheStats]:
+    if adapter_or_others is None:
+        return [CacheStats()]
+
+    if not isinstance(adapter_or_others, BlockAdapter):
+        if not isinstance(adapter_or_others, DiffusionPipeline):
+            transformer = adapter_or_others
+            transformer_2 = None
+        else:
+            transformer = adapter_or_others.transformer
+            transformer_2 = None
+            if hasattr(adapter_or_others, "transformer_2"):
+                transformer_2 = adapter_or_others.transformer_2
+
+        blocks_stats: List[CacheStats] = []
+        for blocks in BlockAdapter.find_blocks(transformer):
+            blocks_stats.append(
+                _summary(
+                    blocks,
+                    details=details,
+                    logging=logging,
+                    **kwargs,
+                )
+            )
+
+        if transformer_2 is not None:
+            for blocks in BlockAdapter.find_blocks(transformer_2):
+                blocks_stats.append(
+                    _summary(
+                        blocks,
+                        details=details,
+                        logging=logging,
+                        **kwargs,
+                    )
+                )
+
+        blocks_stats.append(
+            _summary(
+                transformer,
+                details=details,
+                logging=logging,
+                **kwargs,
+            )
+        )
+        if transformer_2 is not None:
+            blocks_stats.append(
+                _summary(
+                    transformer_2,
+                    details=details,
+                    logging=logging,
+                    **kwargs,
+                )
+            )
+
+        blocks_stats = [stats for stats in blocks_stats if stats.cache_options]
+
+        return blocks_stats if len(blocks_stats) else [CacheStats()]
+
+    adapter = adapter_or_others
+    if not BlockAdapter.check_block_adapter(adapter):
+        return [CacheStats()]
+
+    blocks_stats = []
+    flatten_blocks = BlockAdapter.flatten(adapter.blocks)
+    for blocks in flatten_blocks:
+        blocks_stats.append(
+            _summary(
+                blocks,
+                details=details,
+                logging=logging,
+                **kwargs,
+            )
+        )
+
+    blocks_stats = [stats for stats in blocks_stats if stats.cache_options]
+
+    return blocks_stats if len(blocks_stats) else [CacheStats()]
+
+
+def strify(
+    adapter_or_others: Union[
+        BlockAdapter,
+        DiffusionPipeline,
+        CacheStats,
+        List[CacheStats],
+        Dict[str, Any],
+    ],
+) -> str:
+    if isinstance(adapter_or_others, BlockAdapter):
+        stats = summary(adapter_or_others, logging=False)[-1]
+        cache_options = stats.cache_options
+        cached_steps = len(stats.cached_steps)
+    elif isinstance(adapter_or_others, DiffusionPipeline):
+        stats = summary(adapter_or_others, logging=False)[-1]
+        cache_options = stats.cache_options
+        cached_steps = len(stats.cached_steps)
+    elif isinstance(adapter_or_others, CacheStats):
+        stats = adapter_or_others
+        cache_options = stats.cache_options
+        cached_steps = len(stats.cached_steps)
+    elif isinstance(adapter_or_others, list):
+        stats = adapter_or_others[0]
+        cache_options = stats.cache_options
+        cached_steps = len(stats.cached_steps)
+    elif isinstance(adapter_or_others, dict):
+        from cache_dit.cache_factory import CacheType
+
+        # Assume cache_context_kwargs
+        cache_options = adapter_or_others
+        cached_steps = None
+        cache_type = cache_options.get("cache_type", CacheType.NONE)
+
+        if cache_type == CacheType.NONE:
+            return "NONE"
+    else:
+        raise ValueError(
+            "Please set pipe_or_stats param as one of: "
+            "DiffusionPipeline | CacheStats | Dict[str, Any]"
+        )
+
+    if not cache_options:
+        return "NONE"
+
+    def get_taylorseer_order():
+        taylorseer_order = 0
+        if "taylorseer_order" in cache_options:
+            taylorseer_order = cache_options["taylorseer_order"]
+        return taylorseer_order
+
+    cache_type_str = (
+        f"DBCACHE_F{cache_options.get('Fn_compute_blocks', 1)}"
+        f"B{cache_options.get('Bn_compute_blocks', 0)}_"
+        f"W{cache_options.get('max_warmup_steps', 0)}"
+        f"M{max(0, cache_options.get('max_cached_steps', -1))}"
+        f"MC{max(0, cache_options.get('max_continuous_cached_steps', -1))}_"
+        f"T{int(cache_options.get('enable_taylorseer', False))}"
+        f"O{get_taylorseer_order()}_"
+        f"R{cache_options.get('residual_diff_threshold', 0.08)}"
+    )
+
+    if cached_steps:
+        cache_type_str += f"_S{cached_steps}"
+
+    return cache_type_str
+
+
+def _summary(
+    pipe_or_module: Union[
+        DiffusionPipeline,
+        torch.nn.Module,
+    ],
+    details: bool = False,
+    logging: bool = True,
+    **kwargs,
 ) -> CacheStats:
     cache_stats = CacheStats()
 
@@ -52,6 +211,8 @@ def summary(
         cache_stats.cache_options = cache_options
         if logging:
             print(f"\n🤗Cache Options: {cls_name}\n\n{cache_options}")
+    else:
+        logger.warning(f"Can't find Cache Options for: {cls_name}")
 
     if hasattr(module, "_cached_steps"):
         cached_steps: list[int] = module._cached_steps
@@ -142,58 +303,3 @@ def summary(
                 )
 
     return cache_stats
-
-
-def strify(
-    pipe_or_stats: DiffusionPipeline | CacheStats | Dict[str, Any],
-) -> str:
-    if isinstance(pipe_or_stats, DiffusionPipeline):
-        stats = summary(pipe_or_stats, logging=False)
-        cache_options = stats.cache_options
-        cached_steps = len(stats.cached_steps)
-    elif isinstance(pipe_or_stats, CacheStats):
-        stats = pipe_or_stats
-        cache_options = stats.cache_options
-        cached_steps = len(stats.cached_steps)
-    elif isinstance(pipe_or_stats, dict):
-        # Assume cache_context_kwargs
-        cache_options = pipe_or_stats
-        cached_steps = None
-    else:
-        raise ValueError(
-            "Please set pipe_or_stats param as one of: "
-            "DiffusionPipeline | CacheStats | Dict[str, Any]"
-        )
-
-    if not cache_options:
-        return "NONE"
-
-    if cache_options.get("cache_type", None) != CacheType.DBCache:
-        return "NONE"
-
-    def get_taylorseer_order():
-        taylorseer_order = 0
-        if "taylorseer_kwargs" in cache_options:
-            if "n_derivatives" in cache_options["taylorseer_kwargs"]:
-                taylorseer_order = cache_options["taylorseer_kwargs"][
-                    "n_derivatives"
-                ]
-        elif "taylorseer_order" in cache_options:
-            taylorseer_order = cache_options["taylorseer_order"]
-        return taylorseer_order
-
-    cache_type_str = (
-        f"DBCACHE_F{cache_options.get('Fn_compute_blocks', 1)}"
-        f"B{cache_options.get('Bn_compute_blocks', 0)}_"
-        f"W{cache_options.get('max_warmup_steps', 0)}"
-        f"M{max(0, cache_options.get('max_cached_steps', -1))}"
-        f"MC{max(0, cache_options.get('max_continuous_cached_steps', -1))}_"
-        f"T{int(cache_options.get('enable_taylorseer', False))}"
-        f"O{get_taylorseer_order()}_"
-        f"R{cache_options.get('residual_diff_threshold', 0.08)}"
-    )
-
-    if cached_steps:
-        cache_type_str += f"_S{cached_steps}"
-
-    return cache_type_str

{cache_dit-0.2.27.dist-info → cache_dit-0.2.29.dist-info}/METADATA

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: cache_dit
-Version: 0.2.27
-Summary: 🤗 CacheDiT: An Unified and Training-free Cache Acceleration Toolbox for Diffusion Transformers
+Version: 0.2.29
+Summary: 🤗 A Unified and Training-free Cache Acceleration Toolbox for Diffusion Transformers
 Author: DefTruth, vipshop.com, etc.
 Maintainer: DefTruth, vipshop.com, etc
 Project-URL: Repository, https://github.com/vipshop/cache-dit.git
@@ -43,8 +43,8 @@ Dynamic: requires-python
 <div align="center">
   <img src=https://github.com/vipshop/cache-dit/raw/main/assets/cache-dit-logo.png height="120">
 
-  <p align="center">
-    An <b>Unified</b> and Training-free <b>Cache Acceleration</b> Toolbox for <b>Diffusion Transformers</b> <br>
+<p align="center">
+    A <b>Unified</b> and Training-free <b>Cache Acceleration</b> Toolbox for <b>Diffusion Transformers</b> <br>
     ♥️ <b>Cache Acceleration</b> with <b>One-line</b> Code ~ ♥️
   </p>
   <div align='center'>
@@ -59,32 +59,39 @@ Dynamic: requires-python
   🔥<b><a href="#unified">Unified Cache APIs</a> | <a href="#dbcache">DBCache</a> | <a href="#taylorseer">Hybrid TaylorSeer</a> | <a href="#cfg">Hybrid Cache CFG</a></b>🔥
   </p>
   <p align="center">
-  🎉Now, <b>cache-dit</b> covers <b>All</b> mainstream <b>DiT-based</b> Diffusers' Pipelines</b>🎉<br>
+  🎉Now, <b>cache-dit</b> covers <b>100%</b> Diffusers' <b>DiT-based</b> Pipelines🎉<br>
   🔥<b><a href="#supported">Qwen-Image</a> | <a href="#supported">FLUX.1</a> | <a href="#supported">Wan 2.1/2.2</a> | <a href="#supported"> ... </a> | <a href="#supported">CogVideoX</a></b>🔥
   </p>
 </div>
+<div align='center'>
+  <img src=./assets/gifs/wan2.2.C0_Q0_NONE.gif width=160px>
+  <img src=./assets/gifs/wan2.2.C1_Q0_DBCACHE_F1B0_W2M8MC2_T1O2_R0.08.gif width=160px>
+  <img src=./assets/gifs/wan2.2.C1_Q1_fp8_w8a8_dq_DBCACHE_F1B0_W2M8MC2_T1O2_R0.08.gif width=160px>
+  <p><b>🔥Wan2.2 MoE</b> Baseline | <b><a href="https://github.com/vipshop/cache-dit">+cache-dit</a>:~2.0x↑🎉</b> | +FP8 DQ:<b>~2.4x↑🎉</b></p>
+  <img src=./assets/qwen-image.C0_Q0_NONE.png width=160px>
+  <img src=./assets/qwen-image.C1_Q0_DBCACHE_F8B0_W8M0MC0_T1O4_R0.12_S23.png width=160px>
+  <img src=./assets/qwen-image.C1_Q1_fp8_w8a8_dq_DBCACHE_F8B0_W8M0MC0_T1O4_R0.12_S18.png width=160px>
+  <p><b>🔥Qwen-Image</b> Baseline | <b><a href="https://github.com/vipshop/cache-dit">+cache-dit</a>:~1.8x↑🎉</b> | +FP8 DQ:<b>~2.2x↑🎉</b><br>♥️ Please consider to leave a <b>⭐️ Star</b> to support us ~ ♥️</p>
+  </p>
+</div>
 
 ## 🔥News  
 
-- [2025-09-01] 📚[**Hybird Forward Pattern**](#unified) is supported! Please check [FLUX.1-dev](./examples/run_flux_adapter.py) as an example.
-- [2025-08-29] 🔥</b>Covers <b>All</b> Diffusers' <b>DiT-based</b> Pipelines via **[BlockAdapter](#unified) + [Pattern Matching](#unified).**
-- [2025-08-26] 🎉[**Wan2.2**](https://github.com/Wan-Video) **1.8x⚡️** speedup with `cache-dit + compile`! Please check the [example](./examples/run_wan_2.2.py).
-- [2025-08-19] 🔥[**Qwen-Image-Edit**](https://github.com/QwenLM/Qwen-Image) **2x⚡️** speedup! Check the example at [run_qwen_image_edit.py](./examples/run_qwen_image_edit.py).
+- [2025-09-03] 🎉[**Wan2.2-MoE**](https://github.com/Wan-Video) **2.4x↑🎉** speedup! Please refer to [run_wan_2.2.py](./examples/pipeline/run_wan_2.2.py) as an example.
+- [2025-08-19] 🔥[**Qwen-Image-Edit**](https://github.com/QwenLM/Qwen-Image) **2x↑🎉** speedup! Check the example: [run_qwen_image_edit.py](./examples/pipeline/run_qwen_image_edit.py).
 - [2025-08-12] 🎉First caching mechanism in [QwenLM/Qwen-Image](https://github.com/QwenLM/Qwen-Image) with **[cache-dit](https://github.com/vipshop/cache-dit)**, check this [PR](https://github.com/QwenLM/Qwen-Image/pull/61). 
-- [2025-08-11] 🔥[**Qwen-Image**](https://github.com/QwenLM/Qwen-Image) **1.8x⚡️** speedup! Please refer [run_qwen_image.py](./examples/run_qwen_image.py) as an example.
-- [2025-08-10] 🔥[**FLUX.1-Kontext-dev**](https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev) is supported! Please refer [run_flux_kontext.py](./examples/run_flux_kontext.py) as an example.
-- [2025-07-18] 🎉First caching mechanism in [🤗huggingface/flux-fast](https://github.com/huggingface/flux-fast) with **[cache-dit](https://github.com/vipshop/cache-dit)**, check the [PR](https://github.com/huggingface/flux-fast/pull/13). 
-- [2025-07-13] **[🤗flux-faster](https://github.com/xlite-dev/flux-faster)** is released! **3.3x** speedup for FLUX.1 on NVIDIA L20 with **[cache-dit](https://github.com/vipshop/cache-dit)**.
+- [2025-08-11] 🔥[**Qwen-Image**](https://github.com/QwenLM/Qwen-Image) **1.8x↑🎉** speedup! Please refer to [run_qwen_image.py](./examples/pipeline/run_qwen_image.py) as an example.
+- [2025-07-13] 🎉[**FLUX.1-Dev**](https://github.com/xlite-dev/flux-faster) **3.3x↑🎉** speedup! NVIDIA L20 with **[cache-dit](https://github.com/vipshop/cache-dit)** + `compile + FP8 DQ`.
 
-<!--
 <details>
 <summary> Previous News </summary>  
 
+- [2025-09-01] 📚[**Hybird Forward Pattern**](#unified) is supported! Please check [FLUX.1-dev](./examples/run_flux_adapter.py) as an example.
+- [2025-08-29] 🔥</b>Covers <b>100%</b> Diffusers' <b>DiT-based</b> Pipelines: **[BlockAdapter](#unified) + [Pattern Matching](#unified).**  
+- [2025-08-10] 🔥[**FLUX.1-Kontext-dev**](https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev) is supported! Please refer [run_flux_kontext.py](./examples/pipeline/run_flux_kontext.py) as an example.
 - [2025-07-18] 🎉First caching mechanism in [🤗huggingface/flux-fast](https://github.com/huggingface/flux-fast) with **[cache-dit](https://github.com/vipshop/cache-dit)**, check the [PR](https://github.com/huggingface/flux-fast/pull/13). 
-- [2025-07-13] **[🤗flux-faster](https://github.com/xlite-dev/flux-faster)** is released! **3.3x** speedup for FLUX.1 on NVIDIA L20 with **[cache-dit](https://github.com/vipshop/cache-dit)**.
 
 </details>
--->
 
 ## 📖Contents 
 
@@ -93,6 +100,11 @@ Dynamic: requires-python
 - [⚙️Installation](#️installation)
 - [🔥Supported Models](#supported)
 - [🎉Unified Cache APIs](#unified)
+  - [📚Forward Pattern Matching](#unified)
+  - [🎉Cache with One-line Code](#unified)
+  - [🔥Automatic Block Adapter](#unified)
+  - [📚Hybird Forward Pattern](#unified)
+  - [🤖Cache Acceleration Stats](#unified)
 - [⚡️Dual Block Cache](#dbcache)
 - [🔥Hybrid TaylorSeer](#taylorseer)
 - [⚡️Hybrid Cache CFG](#cfg)
@@ -185,7 +197,7 @@ Currently, for any **Diffusion** models with **Transformer Blocks** that match t
 
 ### ♥️Cache Acceleration with One-line Code
 
-In most cases, you only need to call **one-line** of code, that is `cache_dit.enable_cache(...)`. After this API is called, you just need to call the pipe as normal. The `pipe` param can be **any** Diffusion Pipeline. Please refer to [Qwen-Image](./examples/run_qwen_image.py) as an example. 
+In most cases, you only need to call **one-line** of code, that is `cache_dit.enable_cache(...)`. After this API is called, you just need to call the pipe as normal. The `pipe` param can be **any** Diffusion Pipeline. Please refer to [Qwen-Image](./examples/pipeline/run_qwen_image.py) as an example. 
 
 ```python
 import cache_dit
@@ -199,11 +211,14 @@ cache_dit.enable_cache(pipe)
 
 # Just call the pipe as normal.
 output = pipe(...)
+
+# Disable cache and run original pipe.
+cache_dit.disable_cache(pipe)
 ```
 
-### 🔥Automatic Block Adapter + 📚Hybird Forward Pattern
+### 🔥Automatic Block Adapter
 
-But in some cases, you may have a **modified** Diffusion Pipeline or Transformer that is not located in the diffusers library or not officially supported by **cache-dit** at this time. The **BlockAdapter** can help you solve this problems. Please refer to [🔥Qwen-Image w/ BlockAdapter](./examples/run_qwen_image_adapter.py) and [📚FLUX.1 w/ Hybird Forward Pattern](./examples/run_flux_adapter.py) for more details.
+But in some cases, you may have a **modified** Diffusion Pipeline or Transformer that is not located in the diffusers library or not officially supported by **cache-dit** at this time. The **BlockAdapter** can help you solve this problems. Please refer to [🔥Qwen-Image w/ BlockAdapter](./examples/adapter/run_qwen_image_adapter.py) as an example.
 
 ```python
 from cache_dit import ForwardPattern, BlockAdapter
@@ -225,14 +240,19 @@ cache_dit.enable_cache(
         pipe=pipe, # Qwen-Image, etc.
         transformer=pipe.transformer,
         blocks=pipe.transformer.transformer_blocks,
-        blocks_name="transformer_blocks",
         forward_pattern=ForwardPattern.Pattern_1,
     ), 
 )
+```
+For such situations, **BlockAdapter** can help you quickly apply various cache acceleration features to your own Diffusion Pipelines and Transformers. Please check the [📚BlockAdapter.md](./docs/BlockAdapter.md) for more details.
+
+### 📚Hybird Forward Pattern
 
-# cache-dit supported 📚Hybird Forward Pattern, for example:
+Sometimes, a Transformer class will contain more than one transformer `blocks`. For example, **FLUX.1** (HiDream, Chroma, etc) contains transformer_blocks and single_transformer_blocks (with different forward patterns). The **BlockAdapter** can also help you solve this problem. Please refer to [📚FLUX.1](./examples/adapter/run_flux_adapter.py) as an example.
+
+```python
 # For diffusers <= 0.34.0, FLUX.1 transformer_blocks and 
-# single_transformer_blocks has different forward pattern.
+# single_transformer_blocks have different forward patterns.
 cache_dit.enable_cache(
     BlockAdapter(
         pipe=pipe, # FLUX.1, etc.
@@ -241,10 +261,6 @@ cache_dit.enable_cache(
             pipe.transformer.transformer_blocks,
             pipe.transformer.single_transformer_blocks,
         ],
-        blocks_name=[
-            "transformer_blocks",
-            "single_transformer_blocks",
-        ],
         forward_pattern=[
             ForwardPattern.Pattern_1,
             ForwardPattern.Pattern_3,
@@ -252,7 +268,6 @@ cache_dit.enable_cache(
     ),
 )
 ```
-For such situations, **BlockAdapter** can help you quickly apply various cache acceleration features to your own Diffusion Pipelines and Transformers. Please check the [📚BlockAdapter.md](./docs/BlockAdapter.md) for more details.
 
 ### 🤖Cache Acceleration Stats Summary
 
@@ -377,7 +392,7 @@ cache_dit.enable_cache(
 
 <div id="cfg"></div>
 
-cache-dit supports caching for **CFG (classifier-free guidance)**. For models that fuse CFG and non-CFG into a single forward step, or models that do not include CFG (classifier-free guidance) in the forward step, please set `do_separate_cfg` param to **False (default)**. Otherwise, set it to True. For examples:
+cache-dit supports caching for **CFG (classifier-free guidance)**. For models that fuse CFG and non-CFG into a single forward step, or models that do not include CFG (classifier-free guidance) in the forward step, please set `enable_spearate_cfg` param to **False (default)**. Otherwise, set it to True. For examples:
 
 ```python
 cache_dit.enable_cache(
@@ -385,10 +400,10 @@ cache_dit.enable_cache(
     ...,
     # CFG: classifier free guidance or not
     # For model that fused CFG and non-CFG into single forward step,
-    # should set do_separate_cfg as False. For example, set it as True 
+    # should set enable_spearate_cfg as False. For example, set it as True 
     # for Wan 2.1/Qwen-Image and set it as False for FLUX.1, HunyuanVideo, 
     # CogVideoX, Mochi, LTXVideo, Allegro, CogView3Plus, EasyAnimate, SD3, etc.
-    do_separate_cfg=True, # Wan 2.1, Qwen-Image, CogView4, Cosmos, SkyReelsV2, etc.
+    enable_spearate_cfg=True, # Wan 2.1, Qwen-Image, CogView4, Cosmos, SkyReelsV2, etc.
     # Compute cfg forward first or not, default False, namely, 
     # 0, 2, 4, ..., -> non-CFG step; 1, 3, 5, ... -> CFG step.
     cfg_compute_first=False,
@@ -451,11 +466,21 @@ cache-dit-metrics-cli all  -i1 true_dir -i2 test_dir  # image dir
 
 How to contribute? Star ⭐️ this repo to support us or check [CONTRIBUTE.md](./CONTRIBUTE.md).
 
-## ©️License   
+<div align='center'>
+<a href="https://star-history.com/#vipshop/cache-dit&Date">
+  <picture align='center'>
+    <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=vipshop/cache-dit&type=Date&theme=dark" />
+    <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=vipshop/cache-dit&type=Date" />
+    <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=vipshop/cache-dit&type=Date" width=400px />
+  </picture>
+</a>
+</div>
+
+## ©️Acknowledgements
 
-<div id="license"></div>
+<div id="Acknowledgements"></div>
 
-The **cache-dit** codebase is adapted from FBCache. Special thanks to their excellent work! We have followed the original License from FBCache, please check [LICENSE](./LICENSE) for more details.
+The **cache-dit** codebase is adapted from FBCache. Over time its codebase diverged a lot, and **cache-dit** API is no longer compatible with FBCache. 
 
 ## ©️Citations
 
@@ -463,7 +488,7 @@ The **cache-dit** codebase is adapted from FBCache. Special thanks to their exce
 
 ```BibTeX
 @misc{cache-dit@2025,
-  title={cache-dit: An Unified and Training-free Cache Acceleration Toolbox for Diffusion Transformers},
+  title={cache-dit: A Unified and Training-free Cache Acceleration Toolbox for Diffusion Transformers},
   url={https://github.com/vipshop/cache-dit.git},
   note={Open-source software available at https://github.com/vipshop/cache-dit.git},
   author={vipshop.com},