cache-dit 0.1.1__tar.gz → 0.1.2__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (75) hide show
  1. {cache_dit-0.1.1 → cache_dit-0.1.2}/PKG-INFO +28 -24
  2. {cache_dit-0.1.1 → cache_dit-0.1.2}/README.md +24 -20
  3. cache_dit-0.1.2/assets/cache-dit.png +0 -0
  4. {cache_dit-0.1.1 → cache_dit-0.1.2}/pyproject.toml +3 -3
  5. {cache_dit-0.1.1 → cache_dit-0.1.2}/setup.py +1 -1
  6. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/_version.py +2 -2
  7. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit.egg-info/PKG-INFO +28 -24
  8. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit.egg-info/SOURCES.txt +1 -0
  9. {cache_dit-0.1.1 → cache_dit-0.1.2}/.github/workflows/issue.yml +0 -0
  10. {cache_dit-0.1.1 → cache_dit-0.1.2}/.gitignore +0 -0
  11. {cache_dit-0.1.1 → cache_dit-0.1.2}/.pre-commit-config.yaml +0 -0
  12. {cache_dit-0.1.1 → cache_dit-0.1.2}/CONTRIBUTE.md +0 -0
  13. {cache_dit-0.1.1 → cache_dit-0.1.2}/LICENSE +0 -0
  14. {cache_dit-0.1.1 → cache_dit-0.1.2}/MANIFEST.in +0 -0
  15. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F12B12S4_R0.2_S16.png +0 -0
  16. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F12B16S4_R0.08_S6.png +0 -0
  17. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F16B16S2_R0.2_S14.png +0 -0
  18. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F16B16S4_R0.2_S13.png +0 -0
  19. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F1B0S1_R0.08_S11.png +0 -0
  20. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F1B0S1_R0.2_S19.png +0 -0
  21. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F8B0S2_R0.12_S12.png +0 -0
  22. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F8B16S1_R0.2_S18.png +0 -0
  23. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F8B8S1_R0.08_S9.png +0 -0
  24. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F8B8S1_R0.12_S12.png +0 -0
  25. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCACHE_F8B8S1_R0.15_S15.png +0 -0
  26. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCache.png +0 -0
  27. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png +0 -0
  28. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png +0 -0
  29. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png +0 -0
  30. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png +0 -0
  31. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.07_P52.3_T12.53s.png +0 -0
  32. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.08_P52.4_T12.52s.png +0 -0
  33. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.09_P59.2_T10.81s.png +0 -0
  34. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.12_P59.5_T10.76s.png +0 -0
  35. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.12_P63.0_T9.90s.png +0 -0
  36. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.1_P62.8_T9.95s.png +0 -0
  37. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png +0 -0
  38. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBPRUNE_F1B0_R0.3_P63.1_T9.79s.png +0 -0
  39. {cache_dit-0.1.1 → cache_dit-0.1.2}/assets/NONE_R0.08_S0.png +0 -0
  40. {cache_dit-0.1.1 → cache_dit-0.1.2}/bench/.gitignore +0 -0
  41. {cache_dit-0.1.1 → cache_dit-0.1.2}/bench/bench.py +0 -0
  42. {cache_dit-0.1.1 → cache_dit-0.1.2}/docs/.gitignore +0 -0
  43. {cache_dit-0.1.1 → cache_dit-0.1.2}/examples/.gitignore +0 -0
  44. {cache_dit-0.1.1 → cache_dit-0.1.2}/examples/run_flux.py +0 -0
  45. {cache_dit-0.1.1 → cache_dit-0.1.2}/pytest.ini +0 -0
  46. {cache_dit-0.1.1 → cache_dit-0.1.2}/requirements.txt +0 -0
  47. {cache_dit-0.1.1 → cache_dit-0.1.2}/setup.cfg +0 -0
  48. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/__init__.py +0 -0
  49. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/__init__.py +0 -0
  50. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dual_block_cache/__init__.py +0 -0
  51. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dual_block_cache/cache_context.py +0 -0
  52. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters/__init__.py +0 -0
  53. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters/cogvideox.py +0 -0
  54. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters/flux.py +0 -0
  55. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters/mochi.py +0 -0
  56. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dynamic_block_prune/__init__.py +0 -0
  57. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dynamic_block_prune/diffusers_adapters/__init__.py +0 -0
  58. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dynamic_block_prune/diffusers_adapters/cogvideox.py +0 -0
  59. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dynamic_block_prune/diffusers_adapters/flux.py +0 -0
  60. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dynamic_block_prune/diffusers_adapters/mochi.py +0 -0
  61. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/dynamic_block_prune/prune_context.py +0 -0
  62. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/first_block_cache/__init__.py +0 -0
  63. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/first_block_cache/cache_context.py +0 -0
  64. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/first_block_cache/diffusers_adapters/__init__.py +0 -0
  65. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/first_block_cache/diffusers_adapters/cogvideox.py +0 -0
  66. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/first_block_cache/diffusers_adapters/flux.py +0 -0
  67. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/first_block_cache/diffusers_adapters/mochi.py +0 -0
  68. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/first_block_cache/diffusers_adapters/wan.py +0 -0
  69. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/taylorseer.py +0 -0
  70. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/cache_factory/utils.py +0 -0
  71. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/logger.py +0 -0
  72. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/primitives.py +0 -0
  73. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit.egg-info/dependency_links.txt +0 -0
  74. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit.egg-info/requires.txt +0 -0
  75. {cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit.egg-info/top_level.txt +0 -0
{cache_dit-0.1.1 → cache_dit-0.1.2}/PKG-INFO RENAMED
@@ -1,11 +1,11 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: cache_dit
3
- Version: 0.1.1
4
- Summary: ⚡️DBCache: A Training-free UNet-style Cache Acceleration for Diffusion Transformers
3
+ Version: 0.1.2
4
+ Summary: 🤗 CacheDiT: A Training-free and Easy-to-use Cache Acceleration Toolbox for Diffusion Transformers
5
5
  Author: DefTruth, vipshop.com, etc.
6
6
  Maintainer: DefTruth, vipshop.com, etc
7
- Project-URL: Repository, https://github.com/vipshop/DBCache.git
8
- Project-URL: Homepage, https://github.com/vipshop/DBCache.git
7
+ Project-URL: Repository, https://github.com/vipshop/cache-dit.git
8
+ Project-URL: Homepage, https://github.com/vipshop/cache-dit.git
9
9
  Requires-Python: >=3.10
10
10
  Description-Content-Type: text/markdown
11
11
  License-File: LICENSE
@@ -35,18 +35,18 @@ Dynamic: requires-python
35
35
 
36
36
  <div align="center">
37
37
  <p align="center">
38
- <h3>⚡️DBCache: A Training-free UNet-style Cache Acceleration for <br>Diffusion Transformers</h2>
38
+ <h3>🤗 CacheDiT: A Training-free and Easy-to-use Cache Acceleration <br>Toolbox for Diffusion Transformers</h3>
39
39
  </p>
40
- <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCache.png >
40
+ <img src=https://github.com/vipshop/cache-dit/raw/dev/assets/cache-dit.png >
41
41
  <div align='center'>
42
42
  <img src=https://img.shields.io/badge/Language-Python-brightgreen.svg >
43
43
  <img src=https://img.shields.io/badge/PRs-welcome-9cf.svg >
44
44
  <img src=https://img.shields.io/badge/PyPI-pass-brightgreen.svg >
45
45
  <img src=https://img.shields.io/badge/Python-3.10|3.11|3.12-9cf.svg >
46
- <img src=https://img.shields.io/badge/Release-v0.1.1-brightgreen.svg >
46
+ <img src=https://img.shields.io/badge/Release-v0.1.2-brightgreen.svg >
47
47
  </div>
48
48
  <p align="center">
49
- DeepCache requires UNet’s U-shape, but DiT lacks it. Most DiT cache accelerators are complex and not training-free. DBCache builds on FBCache to create a training-free, UNet-style cache accelerator for DiT.
49
+ DeepCache is for UNet not DiT. Most DiT cache speedups are complex and not training-free. CacheDiT provides <br>a series of training-free, UNet-style cache accelerators for DiT: DBCache, DBPrune, FBCache, etc.
50
50
  </p>
51
51
  </div>
52
52
 
@@ -69,7 +69,7 @@ Dynamic: requires-python
69
69
  |Baseline(L20x1)|F1B0 (0.08)|F1B0 (0.20)|F8B8 (0.15)|F12B12 (0.20)|F16B16 (0.20)|
70
70
  |:---:|:---:|:---:|:---:|:---:|:---:|
71
71
  |24.85s|15.59s|8.58s|15.41s|15.11s|17.74s|
72
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F1B0S1_R0.08_S11.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F1B0S1_R0.2_S19.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F8B8S1_R0.15_S15.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F12B12S4_R0.2_S16.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F16B16S4_R0.2_S13.png width=105px>|
72
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F1B0S1_R0.08_S11.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F1B0S1_R0.2_S19.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F8B8S1_R0.15_S15.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F12B12S4_R0.2_S16.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F16B16S4_R0.2_S13.png width=105px>|
73
73
  |**Baseline(L20x1)**|**F1B0 (0.08)**|**F8B8 (0.12)**|**F8B12 (0.20)**|**F8B16 (0.20)**|**F8B20 (0.20)**|
74
74
  |27.85s|6.04s|5.88s|5.77s|6.01s|6.20s|
75
75
  |<img src=https://github.com/user-attachments/assets/70ea57f4-d8f2-415b-8a96-d8315974a5e6 width=105px>|<img src=https://github.com/user-attachments/assets/fc0e1a67-19cc-44aa-bf50-04696e7978a0 width=105px> |<img src=https://github.com/user-attachments/assets/d1434896-628c-436b-95ad-43c085a8629e width=105px>|<img src=https://github.com/user-attachments/assets/aaa42cd2-57de-4c4e-8bfb-913018a8251d width=105px>|<img src=https://github.com/user-attachments/assets/dc0ba2a4-ef7c-436d-8a39-67055deab92f width=105px>|<img src=https://github.com/user-attachments/assets/aede466f-61ed-4256-8df0-fecf8020c5ca width=105px>|
@@ -93,7 +93,7 @@ These case studies demonstrate that even with relatively high thresholds (such a
93
93
  |Baseline(L20x1)|Pruned(24%)|Pruned(35%)|Pruned(38%)|Pruned(45%)|Pruned(60%)|
94
94
  |:---:|:---:|:---:|:---:|:---:|:---:|
95
95
  |24.85s|19.43s|16.82s|15.95s|14.24s|10.66s|
96
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
96
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
97
97
 
98
98
  <div align="center">
99
99
  <p align="center">
@@ -103,13 +103,17 @@ These case studies demonstrate that even with relatively high thresholds (such a
103
103
 
104
104
  Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can easily tap into its **Context Parallelism** features for distributed inference.
105
105
 
106
+ <p align="center">
107
+ ♥️ Please consider to leave a ⭐️ Star to support us ~ ♥️
108
+ </p>
109
+
106
110
  ## ©️Citations
107
111
 
108
112
  ```BibTeX
109
- @misc{DBCache@2025,
110
- title={DBCache: A Training-free UNet-style Cache Acceleration for Diffusion Transformers},
111
- url={https://github.com/vipshop/DBCache.git},
112
- note={Open-source software available at https://github.com/vipshop/DBCache.git},
113
+ @misc{CacheDiT@2025,
114
+ title={CacheDiT: A Training-free and Easy-to-use cache acceleration Toolbox for Diffusion Transformers},
115
+ url={https://github.com/vipshop/cache-dit.git},
116
+ note={Open-source software available at https://github.com/vipshop/cache-dit.git},
113
117
  author={vipshop.com},
114
118
  year={2025}
115
119
  }
@@ -119,7 +123,7 @@ Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works ha
119
123
 
120
124
  <div id="reference"></div>
121
125
 
122
- **DBCache** is built upon **FBCache**. The **DBCache** codebase was adapted from FBCache's implementation at the [ParaAttention](https://github.com/chengzeyi/ParaAttention/tree/main/src/para_attn/first_block_cache). We would like to express our sincere gratitude for this excellent work!
126
+ The **CacheDiT** codebase was adapted from FBCache's implementation at the [ParaAttention](https://github.com/chengzeyi/ParaAttention/tree/main/src/para_attn/first_block_cache). We would like to express our sincere gratitude for this excellent work!
123
127
 
124
128
  ## 📖Contents
125
129
 
@@ -140,7 +144,7 @@ Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works ha
140
144
 
141
145
  <div id="installation"></div>
142
146
 
143
- You can install the stable release of `DBCache` from PyPI:
147
+ You can install the stable release of `cache-dit` from PyPI:
144
148
 
145
149
  ```bash
146
150
  pip3 install cache-dit
@@ -148,7 +152,7 @@ pip3 install cache-dit
148
152
  Or you can install the latest develop version from GitHub:
149
153
 
150
154
  ```bash
151
- pip3 install git+https://github.com/vipshop/DBCache.git
155
+ pip3 install git+https://github.com/vipshop/cache-dit.git
152
156
  ```
153
157
 
154
158
  ## ⚡️DBCache: Dual Block Cache
@@ -270,13 +274,13 @@ apply_cache_on_pipe(pipe, **cache_options)
270
274
  |Baseline(L20x1)|Pruned(24%)|Pruned(35%)|Pruned(38%)|Pruned(45%)|Pruned(60%)|
271
275
  |:---:|:---:|:---:|:---:|:---:|:---:|
272
276
  |24.85s|19.43s|16.82s|15.95s|14.24s|10.66s|
273
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
277
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
274
278
 
275
279
  ## 🎉Context Parallelism
276
280
 
277
281
  <div id="context-parallelism"></div>
278
282
 
279
- DBCache and DBPrune are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can **easily tap into** its **Context Parallelism** features for distributed inference. Firstly, install `para-attn` from PyPI:
283
+ **CacheDiT** are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can **easily tap into** its **Context Parallelism** features for distributed inference. Firstly, install `para-attn` from PyPI:
280
284
 
281
285
  ```bash
282
286
  pip3 install para-attn # or install `para-attn` from sources.
@@ -312,7 +316,7 @@ apply_cache_on_pipe(
312
316
 
313
317
  <div id="compile"></div>
314
318
 
315
- **DBCache** and **DBPrune** are designed to work compatibly with `torch.compile`. For example:
319
+ **CacheDiT** are designed to work compatibly with `torch.compile`. For example:
316
320
 
317
321
  ```python
318
322
  apply_cache_on_pipe(
@@ -321,7 +325,7 @@ apply_cache_on_pipe(
321
325
  # Compile the Transformer module
322
326
  pipe.transformer = torch.compile(pipe.transformer)
323
327
  ```
324
- However, users intending to use DBCache and DBPrune for DiT with **dynamic input shapes** should consider increasing the **recompile** **limit** of `torch._dynamo` to achieve better performance.
328
+ However, users intending to use **CacheDiT** for DiT with **dynamic input shapes** should consider increasing the **recompile** **limit** of `torch._dynamo` to achieve better performance.
325
329
 
326
330
  ```python
327
331
  torch._dynamo.config.recompile_limit = 96 # default is 8
@@ -333,9 +337,9 @@ Otherwise, the recompile_limit error may be triggered, causing the module to fal
333
337
 
334
338
  <div id="supported"></div>
335
339
 
336
- - [🚀FLUX.1](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
337
- - [🚀CogVideoX](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
338
- - [🚀Mochi](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
340
+ - [🚀FLUX.1](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
341
+ - [🚀CogVideoX](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
342
+ - [🚀Mochi](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
339
343
 
340
344
  ## 👋Contribute
341
345
  <div id="contribute"></div>
{cache_dit-0.1.1 → cache_dit-0.1.2}/README.md RENAMED
@@ -1,17 +1,17 @@
1
1
  <div align="center">
2
2
  <p align="center">
3
- <h3>⚡️DBCache: A Training-free UNet-style Cache Acceleration for <br>Diffusion Transformers</h2>
3
+ <h3>🤗 CacheDiT: A Training-free and Easy-to-use Cache Acceleration <br>Toolbox for Diffusion Transformers</h3>
4
4
  </p>
5
- <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCache.png >
5
+ <img src=https://github.com/vipshop/cache-dit/raw/dev/assets/cache-dit.png >
6
6
  <div align='center'>
7
7
  <img src=https://img.shields.io/badge/Language-Python-brightgreen.svg >
8
8
  <img src=https://img.shields.io/badge/PRs-welcome-9cf.svg >
9
9
  <img src=https://img.shields.io/badge/PyPI-pass-brightgreen.svg >
10
10
  <img src=https://img.shields.io/badge/Python-3.10|3.11|3.12-9cf.svg >
11
- <img src=https://img.shields.io/badge/Release-v0.1.1-brightgreen.svg >
11
+ <img src=https://img.shields.io/badge/Release-v0.1.2-brightgreen.svg >
12
12
  </div>
13
13
  <p align="center">
14
- DeepCache requires UNet’s U-shape, but DiT lacks it. Most DiT cache accelerators are complex and not training-free. DBCache builds on FBCache to create a training-free, UNet-style cache accelerator for DiT.
14
+ DeepCache is for UNet not DiT. Most DiT cache speedups are complex and not training-free. CacheDiT provides <br>a series of training-free, UNet-style cache accelerators for DiT: DBCache, DBPrune, FBCache, etc.
15
15
  </p>
16
16
  </div>
17
17
 
@@ -34,7 +34,7 @@
34
34
  |Baseline(L20x1)|F1B0 (0.08)|F1B0 (0.20)|F8B8 (0.15)|F12B12 (0.20)|F16B16 (0.20)|
35
35
  |:---:|:---:|:---:|:---:|:---:|:---:|
36
36
  |24.85s|15.59s|8.58s|15.41s|15.11s|17.74s|
37
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F1B0S1_R0.08_S11.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F1B0S1_R0.2_S19.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F8B8S1_R0.15_S15.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F12B12S4_R0.2_S16.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F16B16S4_R0.2_S13.png width=105px>|
37
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F1B0S1_R0.08_S11.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F1B0S1_R0.2_S19.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F8B8S1_R0.15_S15.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F12B12S4_R0.2_S16.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F16B16S4_R0.2_S13.png width=105px>|
38
38
  |**Baseline(L20x1)**|**F1B0 (0.08)**|**F8B8 (0.12)**|**F8B12 (0.20)**|**F8B16 (0.20)**|**F8B20 (0.20)**|
39
39
  |27.85s|6.04s|5.88s|5.77s|6.01s|6.20s|
40
40
  |<img src=https://github.com/user-attachments/assets/70ea57f4-d8f2-415b-8a96-d8315974a5e6 width=105px>|<img src=https://github.com/user-attachments/assets/fc0e1a67-19cc-44aa-bf50-04696e7978a0 width=105px> |<img src=https://github.com/user-attachments/assets/d1434896-628c-436b-95ad-43c085a8629e width=105px>|<img src=https://github.com/user-attachments/assets/aaa42cd2-57de-4c4e-8bfb-913018a8251d width=105px>|<img src=https://github.com/user-attachments/assets/dc0ba2a4-ef7c-436d-8a39-67055deab92f width=105px>|<img src=https://github.com/user-attachments/assets/aede466f-61ed-4256-8df0-fecf8020c5ca width=105px>|
@@ -58,7 +58,7 @@ These case studies demonstrate that even with relatively high thresholds (such a
58
58
  |Baseline(L20x1)|Pruned(24%)|Pruned(35%)|Pruned(38%)|Pruned(45%)|Pruned(60%)|
59
59
  |:---:|:---:|:---:|:---:|:---:|:---:|
60
60
  |24.85s|19.43s|16.82s|15.95s|14.24s|10.66s|
61
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
61
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
62
62
 
63
63
  <div align="center">
64
64
  <p align="center">
@@ -68,13 +68,17 @@ These case studies demonstrate that even with relatively high thresholds (such a
68
68
 
69
69
  Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can easily tap into its **Context Parallelism** features for distributed inference.
70
70
 
71
+ <p align="center">
72
+ ♥️ Please consider to leave a ⭐️ Star to support us ~ ♥️
73
+ </p>
74
+
71
75
  ## ©️Citations
72
76
 
73
77
  ```BibTeX
74
- @misc{DBCache@2025,
75
- title={DBCache: A Training-free UNet-style Cache Acceleration for Diffusion Transformers},
76
- url={https://github.com/vipshop/DBCache.git},
77
- note={Open-source software available at https://github.com/vipshop/DBCache.git},
78
+ @misc{CacheDiT@2025,
79
+ title={CacheDiT: A Training-free and Easy-to-use cache acceleration Toolbox for Diffusion Transformers},
80
+ url={https://github.com/vipshop/cache-dit.git},
81
+ note={Open-source software available at https://github.com/vipshop/cache-dit.git},
78
82
  author={vipshop.com},
79
83
  year={2025}
80
84
  }
@@ -84,7 +88,7 @@ Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works ha
84
88
 
85
89
  <div id="reference"></div>
86
90
 
87
- **DBCache** is built upon **FBCache**. The **DBCache** codebase was adapted from FBCache's implementation at the [ParaAttention](https://github.com/chengzeyi/ParaAttention/tree/main/src/para_attn/first_block_cache). We would like to express our sincere gratitude for this excellent work!
91
+ The **CacheDiT** codebase was adapted from FBCache's implementation at the [ParaAttention](https://github.com/chengzeyi/ParaAttention/tree/main/src/para_attn/first_block_cache). We would like to express our sincere gratitude for this excellent work!
88
92
 
89
93
  ## 📖Contents
90
94
 
@@ -105,7 +109,7 @@ Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works ha
105
109
 
106
110
  <div id="installation"></div>
107
111
 
108
- You can install the stable release of `DBCache` from PyPI:
112
+ You can install the stable release of `cache-dit` from PyPI:
109
113
 
110
114
  ```bash
111
115
  pip3 install cache-dit
@@ -113,7 +117,7 @@ pip3 install cache-dit
113
117
  Or you can install the latest develop version from GitHub:
114
118
 
115
119
  ```bash
116
- pip3 install git+https://github.com/vipshop/DBCache.git
120
+ pip3 install git+https://github.com/vipshop/cache-dit.git
117
121
  ```
118
122
 
119
123
  ## ⚡️DBCache: Dual Block Cache
@@ -235,13 +239,13 @@ apply_cache_on_pipe(pipe, **cache_options)
235
239
  |Baseline(L20x1)|Pruned(24%)|Pruned(35%)|Pruned(38%)|Pruned(45%)|Pruned(60%)|
236
240
  |:---:|:---:|:---:|:---:|:---:|:---:|
237
241
  |24.85s|19.43s|16.82s|15.95s|14.24s|10.66s|
238
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
242
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
239
243
 
240
244
  ## 🎉Context Parallelism
241
245
 
242
246
  <div id="context-parallelism"></div>
243
247
 
244
- DBCache and DBPrune are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can **easily tap into** its **Context Parallelism** features for distributed inference. Firstly, install `para-attn` from PyPI:
248
+ **CacheDiT** are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can **easily tap into** its **Context Parallelism** features for distributed inference. Firstly, install `para-attn` from PyPI:
245
249
 
246
250
  ```bash
247
251
  pip3 install para-attn # or install `para-attn` from sources.
@@ -277,7 +281,7 @@ apply_cache_on_pipe(
277
281
 
278
282
  <div id="compile"></div>
279
283
 
280
- **DBCache** and **DBPrune** are designed to work compatibly with `torch.compile`. For example:
284
+ **CacheDiT** are designed to work compatibly with `torch.compile`. For example:
281
285
 
282
286
  ```python
283
287
  apply_cache_on_pipe(
@@ -286,7 +290,7 @@ apply_cache_on_pipe(
286
290
  # Compile the Transformer module
287
291
  pipe.transformer = torch.compile(pipe.transformer)
288
292
  ```
289
- However, users intending to use DBCache and DBPrune for DiT with **dynamic input shapes** should consider increasing the **recompile** **limit** of `torch._dynamo` to achieve better performance.
293
+ However, users intending to use **CacheDiT** for DiT with **dynamic input shapes** should consider increasing the **recompile** **limit** of `torch._dynamo` to achieve better performance.
290
294
 
291
295
  ```python
292
296
  torch._dynamo.config.recompile_limit = 96 # default is 8
@@ -298,9 +302,9 @@ Otherwise, the recompile_limit error may be triggered, causing the module to fal
298
302
 
299
303
  <div id="supported"></div>
300
304
 
301
- - [🚀FLUX.1](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
302
- - [🚀CogVideoX](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
303
- - [🚀Mochi](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
305
+ - [🚀FLUX.1](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
306
+ - [🚀CogVideoX](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
307
+ - [🚀Mochi](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
304
308
 
305
309
  ## 👋Contribute
306
310
  <div id="contribute"></div>
cache_dit-0.1.2/assets/cache-dit.png ADDED
Binary file
{cache_dit-0.1.1 → cache_dit-0.1.2}/pyproject.toml RENAMED
@@ -4,14 +4,14 @@ name = "cache_dit"
4
4
  dynamic = ["version", "dependencies", "optional-dependencies"]
5
5
  requires-python = ">=3.10"
6
6
  authors = [{name = "DefTruth, vipshop.com, etc."}]
7
- description = "⚡️DBCache: A Training-free UNet-style Cache Acceleration for Diffusion Transformers"
7
+ description = "🤗 CacheDiT: A Training-free and Easy-to-use Cache Acceleration Toolbox for Diffusion Transformers"
8
8
  maintainers = [{name="DefTruth, vipshop.com, etc"}]
9
9
  readme = "README.md"
10
10
 
11
11
  [project.urls]
12
12
 
13
- Repository = "https://github.com/vipshop/DBCache.git"
14
- Homepage = "https://github.com/vipshop/DBCache.git"
13
+ Repository = "https://github.com/vipshop/cache-dit.git"
14
+ Homepage = "https://github.com/vipshop/cache-dit.git"
15
15
 
16
16
  [build-system]
17
17
 
{cache_dit-0.1.1 → cache_dit-0.1.2}/setup.py RENAMED
@@ -44,7 +44,7 @@ def fetch_requirements():
44
44
 
45
45
  setup(
46
46
  name=PACKAGE_NAME,
47
- description="⚡️DBCache: A Training-free UNet-style Cache Acceleration for Diffusion Transformers",
47
+ description="🤗 CacheDiT: A Training-free and Easy-to-use Cache Acceleration Toolbox for Diffusion Transformers",
48
48
  author="vipshop.com",
49
49
  use_scm_version={
50
50
  "write_to": path.join("src", "cache_dit", "_version.py"),
{cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit/_version.py RENAMED
@@ -17,5 +17,5 @@ __version__: str
17
17
  __version_tuple__: VERSION_TUPLE
18
18
  version_tuple: VERSION_TUPLE
19
19
 
20
- __version__ = version = '0.1.1'
21
- __version_tuple__ = version_tuple = (0, 1, 1)
20
+ __version__ = version = '0.1.2'
21
+ __version_tuple__ = version_tuple = (0, 1, 2)
{cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit.egg-info/PKG-INFO RENAMED
@@ -1,11 +1,11 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: cache_dit
3
- Version: 0.1.1
4
- Summary: ⚡️DBCache: A Training-free UNet-style Cache Acceleration for Diffusion Transformers
3
+ Version: 0.1.2
4
+ Summary: 🤗 CacheDiT: A Training-free and Easy-to-use Cache Acceleration Toolbox for Diffusion Transformers
5
5
  Author: DefTruth, vipshop.com, etc.
6
6
  Maintainer: DefTruth, vipshop.com, etc
7
- Project-URL: Repository, https://github.com/vipshop/DBCache.git
8
- Project-URL: Homepage, https://github.com/vipshop/DBCache.git
7
+ Project-URL: Repository, https://github.com/vipshop/cache-dit.git
8
+ Project-URL: Homepage, https://github.com/vipshop/cache-dit.git
9
9
  Requires-Python: >=3.10
10
10
  Description-Content-Type: text/markdown
11
11
  License-File: LICENSE
@@ -35,18 +35,18 @@ Dynamic: requires-python
35
35
 
36
36
  <div align="center">
37
37
  <p align="center">
38
- <h3>⚡️DBCache: A Training-free UNet-style Cache Acceleration for <br>Diffusion Transformers</h2>
38
+ <h3>🤗 CacheDiT: A Training-free and Easy-to-use Cache Acceleration <br>Toolbox for Diffusion Transformers</h3>
39
39
  </p>
40
- <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCache.png >
40
+ <img src=https://github.com/vipshop/cache-dit/raw/dev/assets/cache-dit.png >
41
41
  <div align='center'>
42
42
  <img src=https://img.shields.io/badge/Language-Python-brightgreen.svg >
43
43
  <img src=https://img.shields.io/badge/PRs-welcome-9cf.svg >
44
44
  <img src=https://img.shields.io/badge/PyPI-pass-brightgreen.svg >
45
45
  <img src=https://img.shields.io/badge/Python-3.10|3.11|3.12-9cf.svg >
46
- <img src=https://img.shields.io/badge/Release-v0.1.1-brightgreen.svg >
46
+ <img src=https://img.shields.io/badge/Release-v0.1.2-brightgreen.svg >
47
47
  </div>
48
48
  <p align="center">
49
- DeepCache requires UNet’s U-shape, but DiT lacks it. Most DiT cache accelerators are complex and not training-free. DBCache builds on FBCache to create a training-free, UNet-style cache accelerator for DiT.
49
+ DeepCache is for UNet not DiT. Most DiT cache speedups are complex and not training-free. CacheDiT provides <br>a series of training-free, UNet-style cache accelerators for DiT: DBCache, DBPrune, FBCache, etc.
50
50
  </p>
51
51
  </div>
52
52
 
@@ -69,7 +69,7 @@ Dynamic: requires-python
69
69
  |Baseline(L20x1)|F1B0 (0.08)|F1B0 (0.20)|F8B8 (0.15)|F12B12 (0.20)|F16B16 (0.20)|
70
70
  |:---:|:---:|:---:|:---:|:---:|:---:|
71
71
  |24.85s|15.59s|8.58s|15.41s|15.11s|17.74s|
72
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F1B0S1_R0.08_S11.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F1B0S1_R0.2_S19.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F8B8S1_R0.15_S15.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F12B12S4_R0.2_S16.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBCACHE_F16B16S4_R0.2_S13.png width=105px>|
72
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F1B0S1_R0.08_S11.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F1B0S1_R0.2_S19.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F8B8S1_R0.15_S15.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F12B12S4_R0.2_S16.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBCACHE_F16B16S4_R0.2_S13.png width=105px>|
73
73
  |**Baseline(L20x1)**|**F1B0 (0.08)**|**F8B8 (0.12)**|**F8B12 (0.20)**|**F8B16 (0.20)**|**F8B20 (0.20)**|
74
74
  |27.85s|6.04s|5.88s|5.77s|6.01s|6.20s|
75
75
  |<img src=https://github.com/user-attachments/assets/70ea57f4-d8f2-415b-8a96-d8315974a5e6 width=105px>|<img src=https://github.com/user-attachments/assets/fc0e1a67-19cc-44aa-bf50-04696e7978a0 width=105px> |<img src=https://github.com/user-attachments/assets/d1434896-628c-436b-95ad-43c085a8629e width=105px>|<img src=https://github.com/user-attachments/assets/aaa42cd2-57de-4c4e-8bfb-913018a8251d width=105px>|<img src=https://github.com/user-attachments/assets/dc0ba2a4-ef7c-436d-8a39-67055deab92f width=105px>|<img src=https://github.com/user-attachments/assets/aede466f-61ed-4256-8df0-fecf8020c5ca width=105px>|
@@ -93,7 +93,7 @@ These case studies demonstrate that even with relatively high thresholds (such a
93
93
  |Baseline(L20x1)|Pruned(24%)|Pruned(35%)|Pruned(38%)|Pruned(45%)|Pruned(60%)|
94
94
  |:---:|:---:|:---:|:---:|:---:|:---:|
95
95
  |24.85s|19.43s|16.82s|15.95s|14.24s|10.66s|
96
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
96
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
97
97
 
98
98
  <div align="center">
99
99
  <p align="center">
@@ -103,13 +103,17 @@ These case studies demonstrate that even with relatively high thresholds (such a
103
103
 
104
104
  Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can easily tap into its **Context Parallelism** features for distributed inference.
105
105
 
106
+ <p align="center">
107
+ ♥️ Please consider to leave a ⭐️ Star to support us ~ ♥️
108
+ </p>
109
+
106
110
  ## ©️Citations
107
111
 
108
112
  ```BibTeX
109
- @misc{DBCache@2025,
110
- title={DBCache: A Training-free UNet-style Cache Acceleration for Diffusion Transformers},
111
- url={https://github.com/vipshop/DBCache.git},
112
- note={Open-source software available at https://github.com/vipshop/DBCache.git},
113
+ @misc{CacheDiT@2025,
114
+ title={CacheDiT: A Training-free and Easy-to-use cache acceleration Toolbox for Diffusion Transformers},
115
+ url={https://github.com/vipshop/cache-dit.git},
116
+ note={Open-source software available at https://github.com/vipshop/cache-dit.git},
113
117
  author={vipshop.com},
114
118
  year={2025}
115
119
  }
@@ -119,7 +123,7 @@ Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works ha
119
123
 
120
124
  <div id="reference"></div>
121
125
 
122
- **DBCache** is built upon **FBCache**. The **DBCache** codebase was adapted from FBCache's implementation at the [ParaAttention](https://github.com/chengzeyi/ParaAttention/tree/main/src/para_attn/first_block_cache). We would like to express our sincere gratitude for this excellent work!
126
+ The **CacheDiT** codebase was adapted from FBCache's implementation at the [ParaAttention](https://github.com/chengzeyi/ParaAttention/tree/main/src/para_attn/first_block_cache). We would like to express our sincere gratitude for this excellent work!
123
127
 
124
128
  ## 📖Contents
125
129
 
@@ -140,7 +144,7 @@ Moreover, both DBCache and DBPrune are **plug-and-play** solutions that works ha
140
144
 
141
145
  <div id="installation"></div>
142
146
 
143
- You can install the stable release of `DBCache` from PyPI:
147
+ You can install the stable release of `cache-dit` from PyPI:
144
148
 
145
149
  ```bash
146
150
  pip3 install cache-dit
@@ -148,7 +152,7 @@ pip3 install cache-dit
148
152
  Or you can install the latest develop version from GitHub:
149
153
 
150
154
  ```bash
151
- pip3 install git+https://github.com/vipshop/DBCache.git
155
+ pip3 install git+https://github.com/vipshop/cache-dit.git
152
156
  ```
153
157
 
154
158
  ## ⚡️DBCache: Dual Block Cache
@@ -270,13 +274,13 @@ apply_cache_on_pipe(pipe, **cache_options)
270
274
  |Baseline(L20x1)|Pruned(24%)|Pruned(35%)|Pruned(38%)|Pruned(45%)|Pruned(60%)|
271
275
  |:---:|:---:|:---:|:---:|:---:|:---:|
272
276
  |24.85s|19.43s|16.82s|15.95s|14.24s|10.66s|
273
- |<img src=https://github.com/vipshop/DBCache/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/DBCache/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
277
+ |<img src=https://github.com/vipshop/cache-dit/raw/main/assets/NONE_R0.08_S0.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.03_P24.0_T19.43s.png width=105px> | <img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.04_P34.6_T16.82s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.05_P38.3_T15.95s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.06_P45.2_T14.24s.png width=105px>|<img src=https://github.com/vipshop/cache-dit/raw/main/assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png width=105px>|
274
278
 
275
279
  ## 🎉Context Parallelism
276
280
 
277
281
  <div id="context-parallelism"></div>
278
282
 
279
- DBCache and DBPrune are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can **easily tap into** its **Context Parallelism** features for distributed inference. Firstly, install `para-attn` from PyPI:
283
+ **CacheDiT** are **plug-and-play** solutions that works hand-in-hand with [ParaAttention](https://github.com/chengzeyi/ParaAttention). Users can **easily tap into** its **Context Parallelism** features for distributed inference. Firstly, install `para-attn` from PyPI:
280
284
 
281
285
  ```bash
282
286
  pip3 install para-attn # or install `para-attn` from sources.
@@ -312,7 +316,7 @@ apply_cache_on_pipe(
312
316
 
313
317
  <div id="compile"></div>
314
318
 
315
- **DBCache** and **DBPrune** are designed to work compatibly with `torch.compile`. For example:
319
+ **CacheDiT** are designed to work compatibly with `torch.compile`. For example:
316
320
 
317
321
  ```python
318
322
  apply_cache_on_pipe(
@@ -321,7 +325,7 @@ apply_cache_on_pipe(
321
325
  # Compile the Transformer module
322
326
  pipe.transformer = torch.compile(pipe.transformer)
323
327
  ```
324
- However, users intending to use DBCache and DBPrune for DiT with **dynamic input shapes** should consider increasing the **recompile** **limit** of `torch._dynamo` to achieve better performance.
328
+ However, users intending to use **CacheDiT** for DiT with **dynamic input shapes** should consider increasing the **recompile** **limit** of `torch._dynamo` to achieve better performance.
325
329
 
326
330
  ```python
327
331
  torch._dynamo.config.recompile_limit = 96 # default is 8
@@ -333,9 +337,9 @@ Otherwise, the recompile_limit error may be triggered, causing the module to fal
333
337
 
334
338
  <div id="supported"></div>
335
339
 
336
- - [🚀FLUX.1](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
337
- - [🚀CogVideoX](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
338
- - [🚀Mochi](https://github.com/vipshop/DBCache/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
340
+ - [🚀FLUX.1](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
341
+ - [🚀CogVideoX](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
342
+ - [🚀Mochi](https://github.com/vipshop/cache-dit/raw/main/src/cache_dit/cache_factory/dual_block_cache/diffusers_adapters)
339
343
 
340
344
  ## 👋Contribute
341
345
  <div id="contribute"></div>
{cache_dit-0.1.1 → cache_dit-0.1.2}/src/cache_dit.egg-info/SOURCES.txt RENAMED
@@ -35,6 +35,7 @@ assets/DBPRUNE_F1B0_R0.1_P62.8_T9.95s.png
35
35
  assets/DBPRUNE_F1B0_R0.2_P59.5_T10.66s.png
36
36
  assets/DBPRUNE_F1B0_R0.3_P63.1_T9.79s.png
37
37
  assets/NONE_R0.08_S0.png
38
+ assets/cache-dit.png
38
39
  bench/.gitignore
39
40
  bench/bench.py
40
41
  docs/.gitignore
{cache_dit-0.1.1 → cache_dit-0.1.2}/.gitignore RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/CONTRIBUTE.md RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/LICENSE RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/MANIFEST.in RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/assets/DBCache.png RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/bench/.gitignore RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/bench/bench.py RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/docs/.gitignore RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/examples/.gitignore RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/pytest.ini RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/requirements.txt RENAMED
File without changes
{cache_dit-0.1.1 → cache_dit-0.1.2}/setup.cfg RENAMED
File without changes