cache-dit 0.2.23__py3-none-any.whl → 0.2.25__py3-none-any.whl

cache_dit/quantize/quantize_ao.py

@@ -0,0 +1,182 @@
+import gc
+import time
+import torch
+from typing import Callable, Optional, List
+from cache_dit.logger import init_logger
+
+logger = init_logger(__name__)
+
+
+def quantize_ao(
+    module: torch.nn.Module,
+    quant_type: str = "fp8_w8a8_dq",
+    per_row: bool = True,
+    exclude_layers: List[str] = [
+        "embedder",
+        "embed",
+    ],
+    filter_fn: Optional[Callable] = None,
+    **kwargs,
+) -> torch.nn.Module:
+    # Apply FP8 DQ for module and skip any `embed` modules
+    # by default to avoid non-trivial precision downgrade. Please
+    # set `exclude_layers` as `[]` if you don't want this behavior.
+    assert isinstance(module, torch.nn.Module)
+
+    quant_type = quant_type.lower()
+    assert quant_type in (
+        "fp8_w8a8_dq",
+        "fp8_w8a16_wo",
+        "int8_w8a8_dq",
+        "int8_w8a16_wo",
+        "int4_w4a8_dq",
+        "int4_w4a4_dq",
+        "int4_w4a16_wo",
+    ), f"{quant_type} is not supported for torchao backend now!"
+
+    if "fp8" in quant_type:
+        assert torch.cuda.get_device_capability() >= (
+            8,
+            9,
+        ), "FP8 is not supported for current device."
+
+    num_quant_linear = 0
+    num_skip_linear = 0
+    num_linear_layers = 0
+    num_layers = 0
+
+    # Ensure bfloat16 for per_row
+    def _filter_fn(m: torch.nn.Module, name: str) -> bool:
+        nonlocal num_quant_linear, num_skip_linear, num_linear_layers, num_layers
+        num_layers += 1
+        if isinstance(m, torch.nn.Linear):
+            num_linear_layers += 1
+            for exclude_name in exclude_layers:
+                if exclude_name in name:
+                    logger.info(
+                        f"Skip Quantization: {name} -> "
+                        f"pattern<{exclude_name}>"
+                    )
+
+                    num_skip_linear += 1
+                    return False
+
+            if (
+                per_row
+                and m.weight.dtype != torch.bfloat16
+                and quant_type == "fp8_w8a8_dq"
+            ):
+                logger.info(
+                    f"Skip Quantization: {name} -> "
+                    f"pattern<dtype({m.weight.dtype})!=bfloat16>"
+                )
+
+                num_skip_linear += 1
+                return False
+
+            num_quant_linear += 1
+            return True
+
+        return False
+
+    def _quantization_fn():
+        try:
+            if quant_type == "fp8_w8a8_dq":
+                from torchao.quantization import (
+                    float8_dynamic_activation_float8_weight,
+                    PerTensor,
+                    PerRow,
+                )
+
+                quantization_fn = float8_dynamic_activation_float8_weight(
+                    granularity=(
+                        ((PerRow(), PerRow()))
+                        if per_row
+                        else ((PerTensor(), PerTensor()))
+                    )
+                )
+
+            elif quant_type == "fp8_w8a16_wo":
+                from torchao.quantization import float8_weight_only
+
+                quantization_fn = float8_weight_only()
+
+            elif quant_type == "int8_w8a8_dq":
+                from torchao.quantization import (
+                    int8_dynamic_activation_int8_weight,
+                )
+
+                quantization_fn = int8_dynamic_activation_int8_weight()
+
+            elif quant_type == "int8_w8a16_wo":
+                from torchao.quantization import int8_weight_only
+
+                quantization_fn = int8_weight_only(
+                    # group_size is None -> per_channel, else per group
+                    group_size=kwargs.get("group_size", None),
+                )
+
+            elif quant_type == "int4_w4a8_dq":
+                from torchao.quantization import (
+                    int8_dynamic_activation_int4_weight,
+                )
+
+                quantization_fn = int8_dynamic_activation_int4_weight(
+                    group_size=kwargs.get("group_size", 32),
+                )
+
+            elif quant_type == "int4_w4a4_dq":
+                from torchao.quantization import (
+                    int4_dynamic_activation_int4_weight,
+                )
+
+                quantization_fn = int4_dynamic_activation_int4_weight()
+
+            elif quant_type == "int4_w4a16_wo":
+                from torchao.quantization import int4_weight_only
+
+                quantization_fn = int4_weight_only(
+                    group_size=kwargs.get("group_size", 32),
+                )
+
+            else:
+                raise ValueError(
+                    f"quant_type: {quant_type} is not supported now!"
+                )
+
+        except ImportError as e:
+            e.msg += (
+                f"{quant_type} is not supported in torchao backend now! "
+                "Please upgrade the torchao library."
+            )
+            raise e
+
+        return quantization_fn
+
+    from torchao.quantization import quantize_
+
+    quantize_(
+        module,
+        _quantization_fn(),
+        filter_fn=_filter_fn if filter_fn is None else filter_fn,
+        **kwargs,
+    )
+
+    force_empty_cache()
+
+    logger.info(
+        f"Quantized Linear Layers: {num_quant_linear:>5}\n"
+        f"Skipped   Linear Layers: {num_skip_linear:>5}\n"
+        f"Total     Linear Layers: {num_linear_layers:>5}\n"
+        f"Total     (all)  Layers: {num_layers:>5}"
+    )
+    return module
+
+
+def force_empty_cache():
+    time.sleep(1)
+    gc.collect()
+    torch.cuda.empty_cache()
+    time.sleep(1)
+    gc.collect()
+    torch.cuda.empty_cache()

cache_dit/quantize/quantize_interface.py

@@ -0,0 +1,46 @@
+import torch
+from typing import Callable, Optional, List
+from cache_dit.logger import init_logger
+
+logger = init_logger(__name__)
+
+
+def quantize(
+    module: torch.nn.Module,
+    quant_type: str = "fp8_w8a8_dq",
+    backend: str = "ao",
+    # only for fp8_w8a8_dq
+    per_row: bool = True,
+    exclude_layers: List[str] = [
+        "embedder",
+        "embed",
+    ],
+    filter_fn: Optional[Callable] = None,
+    **kwargs,
+) -> torch.nn.Module:
+    assert isinstance(module, torch.nn.Module)
+
+    if backend.lower() in ("ao", "torchao"):
+        from cache_dit.quantize.quantize_ao import quantize_ao
+
+        quant_type = quant_type.lower()
+        assert quant_type in (
+            "fp8_w8a8_dq",
+            "fp8_w8a16_wo",
+            "int8_w8a8_dq",
+            "int8_w8a16_wo",
+            "int4_w4a8_dq",
+            "int4_w4a4_dq",
+            "int4_w4a16_wo",
+        ), f"{quant_type} is not supported for torchao backend now!"
+
+        return quantize_ao(
+            module,
+            quant_type=quant_type,
+            per_row=per_row,
+            exclude_layers=exclude_layers,
+            filter_fn=filter_fn,
+            **kwargs,
+        )
+    else:
+        raise ValueError(f"backend: {backend} is not supported now!")

cache_dit/utils.py

@@ -5,7 +5,10 @@ import numpy as np
 from pprint import pprint
 from diffusers import DiffusionPipeline
 
+from typing import Dict, Any
 from cache_dit.logger import init_logger
+from cache_dit.cache_factory import CacheType
+
 
 logger = init_logger(__name__)
 
@@ -27,22 +30,26 @@ class CacheStats:
 
 
 def summary(
-    pipe: DiffusionPipeline, details: bool = False, logging: bool = True
-):
+    pipe_or_transformer: DiffusionPipeline | torch.nn.Module,
+    details: bool = False,
+    logging: bool = True,
+) -> CacheStats:
     cache_stats = CacheStats()
-    pipe_cls_name = pipe.__class__.__name__
+    cls_name = pipe_or_transformer.__class__.__name__
+    if isinstance(pipe_or_transformer, DiffusionPipeline):
+        transformer = pipe_or_transformer.transformer
+    else:
+        transformer = pipe_or_transformer
 
-    if hasattr(pipe, "_cache_options"):
-        cache_options = pipe._cache_options
+    if hasattr(transformer, "_cache_context_kwargs"):
+        cache_options = transformer._cache_context_kwargs
         cache_stats.cache_options = cache_options
         if logging:
-            print(f"\n🤗Cache Options: {pipe_cls_name}\n\n{cache_options}")
+            print(f"\n🤗Cache Options: {cls_name}\n\n{cache_options}")
 
-    if hasattr(pipe.transformer, "_cached_steps"):
-        cached_steps: list[int] = pipe.transformer._cached_steps
-        residual_diffs: dict[str, float] = dict(
-            pipe.transformer._residual_diffs
-        )
+    if hasattr(transformer, "_cached_steps"):
+        cached_steps: list[int] = transformer._cached_steps
+        residual_diffs: dict[str, float] = dict(transformer._residual_diffs)
         cache_stats.cached_steps = cached_steps
         cache_stats.residual_diffs = residual_diffs
 
@@ -57,7 +64,7 @@ def summary(
             qmax = np.max(diffs_values)
 
             print(
-                f"\n⚡️Cache Steps and Residual Diffs Statistics: {pipe_cls_name}\n"
+                f"\n⚡️Cache Steps and Residual Diffs Statistics: {cls_name}\n"
             )
 
             print(
@@ -74,9 +81,7 @@ def summary(
             print("")
 
             if details:
-                print(
-                    f"📚Cache Steps and Residual Diffs Details: {pipe_cls_name}\n"
-                )
+                print(f"📚Cache Steps and Residual Diffs Details: {cls_name}\n")
                 pprint(
                     f"Cache Steps: {len(cached_steps)}, {cached_steps}",
                 )
@@ -85,10 +90,10 @@ def summary(
                     compact=True,
                 )
 
-    if hasattr(pipe.transformer, "_cfg_cached_steps"):
-        cfg_cached_steps: list[int] = pipe.transformer._cfg_cached_steps
+    if hasattr(transformer, "_cfg_cached_steps"):
+        cfg_cached_steps: list[int] = transformer._cfg_cached_steps
         cfg_residual_diffs: dict[str, float] = dict(
-            pipe.transformer._cfg_residual_diffs
+            transformer._cfg_residual_diffs
         )
         cache_stats.cfg_cached_steps = cfg_cached_steps
         cache_stats.cfg_residual_diffs = cfg_residual_diffs
@@ -104,7 +109,7 @@ def summary(
             qmax = np.max(cfg_diffs_values)
 
             print(
-                f"\n⚡️CFG Cache Steps and Residual Diffs Statistics: {pipe_cls_name}\n"
+                f"\n⚡️CFG Cache Steps and Residual Diffs Statistics: {cls_name}\n"
             )
 
             print(
@@ -122,7 +127,7 @@ def summary(
 
             if details:
                 print(
-                    f"📚CFG Cache Steps and Residual Diffs Details: {pipe_cls_name}\n"
+                    f"📚CFG Cache Steps and Residual Diffs Details: {cls_name}\n"
                 )
                 pprint(
                     f"CFG Cache Steps: {len(cfg_cached_steps)}, {cfg_cached_steps}",
@@ -135,27 +140,56 @@ def summary(
     return cache_stats
 
 
-def strify(pipe_or_stats: DiffusionPipeline | CacheStats):
-    if not isinstance(pipe_or_stats, CacheStats):
+def strify(
+    pipe_or_stats: DiffusionPipeline | CacheStats | Dict[str, Any],
+) -> str:
+    if isinstance(pipe_or_stats, DiffusionPipeline):
         stats = summary(pipe_or_stats, logging=False)
-    else:
+        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)
+        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['Fn_compute_blocks']}"
-        f"B{cache_options['Bn_compute_blocks']}"
-        f"W{cache_options['warmup_steps']}"
-        f"M{max(0, cache_options['max_cached_steps'])}"
-        f"T{int(cache_options['enable_taylorseer'])}"
-        f"O{cache_options['taylorseer_kwargs']['n_derivatives']}_"
-        f"R{cache_options['residual_diff_threshold']}_"
-        f"S{cached_steps}"  # skiped steps
+        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-0.2.23.dist-info → cache_dit-0.2.25.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cache_dit
-Version: 0.2.23
+Version: 0.2.25
 Summary: 🤗 CacheDiT: An Unified and Training-free Cache Acceleration Toolbox for Diffusion Transformers
 Author: DefTruth, vipshop.com, etc.
 Maintainer: DefTruth, vipshop.com, etc
@@ -11,12 +11,13 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: packaging
 Requires-Dist: pyyaml
-Requires-Dist: torch>=2.5.1
-Requires-Dist: transformers>=4.51.3
-Requires-Dist: diffusers>=0.33.1
+Requires-Dist: torch>=2.7.1
+Requires-Dist: transformers>=4.55.2
+Requires-Dist: diffusers>=0.35.1
 Requires-Dist: scikit-image
 Requires-Dist: scipy
 Requires-Dist: lpips==0.1.4
+Requires-Dist: torchao>=0.12.0
 Provides-Extra: all
 Provides-Extra: dev
 Requires-Dist: pre-commit; extra == "dev"
@@ -59,12 +60,13 @@ Dynamic: requires-python
   </p>
   <p align="center">
   🎉Now, <b>cache-dit</b> covers <b>Most</b> mainstream <b>Diffusers'</b> Pipelines</b>🎉<br>
-  🔥<b><a href="#supported">Qwen-Image</a> | <a href="#supported">FLUX.1</a> | <a href="#supported">Wan 2.1</a> | <a href="#supported"> ... </a> | <a href="#supported">CogVideoX</a></b>🔥
+  🔥<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>
 
 ## 🔥News  
 
+- [2025-08-26] 🎉[**Wan2.2**](https://github.com/Wan-Video) **1.8x⚡️** speedup with `cache-dit + compile`! Check the [example](./examples/run_wan_2.2.py).
 - [2025-08-19] 🔥[**Qwen-Image-Edit**](https://github.com/QwenLM/Qwen-Image) **2x⚡️** speedup! Check example [run_qwen_image_edit.py](./examples/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 the [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.
@@ -119,6 +121,7 @@ Currently, **cache-dit** library supports almost **Any** Diffusion Transformers
 - [🚀FLUX.1-Kontext-dev](https://github.com/vipshop/cache-dit/raw/main/examples)  
 - [🚀CogVideoX](https://github.com/vipshop/cache-dit/raw/main/examples)
 - [🚀CogVideoX1.5](https://github.com/vipshop/cache-dit/raw/main/examples)
+- [🚀Wan2.2-T2V](https://github.com/vipshop/cache-dit/raw/main/examples)
 - [🚀Wan2.1-T2V](https://github.com/vipshop/cache-dit/raw/main/examples)
 - [🚀Wan2.1-FLF2V](https://github.com/vipshop/cache-dit/raw/main/examples)
 - [🚀HunyuanVideo](https://github.com/vipshop/cache-dit/raw/main/examples)
@@ -166,7 +169,7 @@ cache_dit.enable_cache(pipe)
 output = pipe(...)
 ```
 
-### 🔥BlockAdapter: Cache Acceleration for Custom Diffusion Models
+### 🔥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) as an example.
 
@@ -181,7 +184,7 @@ cache_dit.enable_cache(
     forward_pattern=ForwardPattern.Pattern_1,  
 )
 
-# Or, manualy setup transformer configurations.
+# Or, manually setup transformer configurations.
 cache_dit.enable_cache(
     BlockAdapter(
         pipe=pipe, # Qwen-Image, etc.
@@ -238,7 +241,7 @@ cache_dit.enable_cache(pipe)
 # Custom options, F8B8, higher precision
 cache_dit.enable_cache(
     pipe,
-    warmup_steps=8,      # steps do not cache
+    max_warmup_steps=8,  # steps do not cache
     max_cached_steps=-1, # -1 means no limit
     Fn_compute_blocks=8, # Fn, F8, etc.
     Bn_compute_blocks=8, # Bn, B8, etc.
@@ -292,14 +295,11 @@ cache_dit.enable_cache(
     enable_encoder_taylorseer=True,
     # Taylorseer cache type cache be hidden_states or residual.
     taylorseer_cache_type="residual",
-    # Higher values of n_derivatives will lead to longer 
-    # computation time but may improve precision significantly.
-    taylorseer_kwargs={
-        "n_derivatives": 2, # default is 2.
-    },
-    warmup_steps=3, # prefer: >= n_derivatives + 1
+    # Higher values of order will lead to longer computation time
+    taylorseer_order=2, # default is 2.
+    max_warmup_steps=3, # prefer: >= order + 1
     residual_diff_threshold=0.12
-)
+)s
 ``` 
 
 > [!Important]

{cache_dit-0.2.23.dist-info → cache_dit-0.2.25.dist-info}/RECORD

@@ -1,22 +1,21 @@
-cache_dit/__init__.py,sha256=KwhX9NfYkWSvDFuuUVeVjcuiZiGS_22y386l8j4afMo,905
-cache_dit/_version.py,sha256=6GZdGbiFdhndXqR5oFLOd8VGzUvRkESP-NyStAZWYUw,706
+cache_dit/__init__.py,sha256=VsT0f0R0COp8v6Sx9hGNsqxiElERaDpfG11a9MfK0is,945
+cache_dit/_version.py,sha256=t9iixyDuMWz1nP7KM0bgrLNIpwu8JK6uZApA8DoCQwM,706
 cache_dit/logger.py,sha256=0zsu42hN-3-rgGC_C29ms1IvVpV4_b4_SwJCKSenxBE,4304
-cache_dit/primitives.py,sha256=A2iG9YLot3gOsZSPp-_gyjqjLgJvWQRx8aitD4JQ23Y,3877
-cache_dit/utils.py,sha256=3UgVhfmTFG28w6CV-Rfxp5u1uzLrRozocHwLCTGiQ5M,5865
+cache_dit/utils.py,sha256=1oWDMYs6E7FRsd8cidsVOPT-meIRKeuqbGbE6CrCUec,7236
 cache_dit/cache_factory/.gitignore,sha256=5Cb-qT9wsTUoMJ7vACDF7ZcLpAXhi5v-xdcWSRit988,23
 cache_dit/cache_factory/__init__.py,sha256=evWenCin1kuBGa6W5BCKMrDZc1C1R2uVPSg0BjXgdXE,499
-cache_dit/cache_factory/cache_adapters.py,sha256=QTSwjdCmHDeF80TLp6D3KhzQS_oMPna0_bESgJBrdkg,23978
+cache_dit/cache_factory/cache_adapters.py,sha256=Yugqljm9tm615srM2BGQlR_tA0QiZo3PbLPceObh4dQ,25988
 cache_dit/cache_factory/cache_blocks.py,sha256=ZeazBsYvLIjI5M_OnLL2xP2W7zMeM0rxVfBBwIVHBRs,18661
-cache_dit/cache_factory/cache_context.py,sha256=4thx9NYxVaYZ_Nr2quUVE8bsNmTsXhZK0F960rccOc8,39000
-cache_dit/cache_factory/cache_interface.py,sha256=PohG_2oy747O37YSsWz_DwxxTXN7ORhQatyEbg_6fQs,8045
+cache_dit/cache_factory/cache_context.py,sha256=HhA5IMSdF-i-koSB1jqf5AMC_UyDV7VinpHm4Qee9Ig,41800
+cache_dit/cache_factory/cache_interface.py,sha256=HymagnKEDs48Ly_x3IM5jTMNJpLrIdJnppVlkr2xHaM,8433
 cache_dit/cache_factory/cache_types.py,sha256=FIFa6ZBfvvSMMHyBBhvarvgg2Y2wbRgITcG_uGylGe0,991
 cache_dit/cache_factory/forward_pattern.py,sha256=B2YeqV2t_zo2Ar8m7qimPBjwQgoXHGp2grPZmEAhi8s,1286
-cache_dit/cache_factory/taylorseer.py,sha256=WeK2WlAJa4Px_pnAKokmnZXeqQYylQkPw4-EDqBIqeQ,3770
-cache_dit/cache_factory/utils.py,sha256=YGtn02O3fVlrfQ32gGV4WAtTRvzzwSXNxzP_FmnE2Uk,1867
+cache_dit/cache_factory/taylorseer.py,sha256=etSUIZzDvqW3ScKCbccTPcFaSmxV1T-xAXdk-p3e3wk,3802
+cache_dit/cache_factory/utils.py,sha256=XkVM9AXcB9zYq8-S8QKAsGz80r3tA6U3lBNGDGeHOe4,1871
 cache_dit/cache_factory/patch/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 cache_dit/cache_factory/patch/flux.py,sha256=iNQ-1RlOgXupZ4uPiEvJ__Ro6vKT_fOKja9JrpMrO78,8998
 cache_dit/compile/__init__.py,sha256=FcTVzCeyypl-mxlc59_ehHL3lBNiDAFsXuRoJ-5Cfi0,56
-cache_dit/compile/utils.py,sha256=53KPsMWHyGmHGtw0T4oP0VY4O60cVXOFwFGWTlZrUqI,3857
+cache_dit/compile/utils.py,sha256=nN2OIrSdwRR5zGxJinKDqb07pXpvTNTF3g_OgLkeeBU,3858
 cache_dit/custom_ops/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 cache_dit/custom_ops/triton_taylorseer.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 cache_dit/metrics/__init__.py,sha256=RaUhl5dieF40RqnizGzR30qoJJ9dyMUEADwgwMaMQrE,575
@@ -25,9 +24,13 @@ cache_dit/metrics/fid.py,sha256=9Ivtazl6mW0Bon2VXa-Ia5Xj2ewxRD3V1Qkd69zYM3Y,1706
 cache_dit/metrics/inception.py,sha256=pBVe2X6ylLPIXTG4-GWDM9DWnCviMJbJ45R3ulhktR0,12759
 cache_dit/metrics/lpips.py,sha256=I2qCNi6qJh5TRsaIsdxO0WoRX1DN7U_H3zS0oCSahYM,1032
 cache_dit/metrics/metrics.py,sha256=8jvM1sF-nDxUuwCRy44QEoo4dYVLCQVh1QyAMs4eaQY,27840
-cache_dit-0.2.23.dist-info/licenses/LICENSE,sha256=Dqb07Ik2dV41s9nIdMUbiRWEfDqo7-dQeRiY7kPO8PE,3769
-cache_dit-0.2.23.dist-info/METADATA,sha256=Dq2f8TlyTmv36otIJ2F-fRGkJlZmpW2SY6O14P2AYKo,19772
-cache_dit-0.2.23.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-cache_dit-0.2.23.dist-info/entry_points.txt,sha256=FX2gysXaZx6NeK1iCLMcIdP8Q4_qikkIHtEmi3oWn8o,65
-cache_dit-0.2.23.dist-info/top_level.txt,sha256=ZJDydonLEhujzz0FOkVbO-BqfzO9d_VqRHmZU-3MOZo,10
-cache_dit-0.2.23.dist-info/RECORD,,
+cache_dit/quantize/__init__.py,sha256=kWYoMAyZgBXu9BJlZjTQ0dRffW9GqeeY9_iTkXrb70A,59
+cache_dit/quantize/quantize_ao.py,sha256=sKz_RmVtxLOpAPnUv_iOjzY_226pfaxgB_HMNrfyqB8,5465
+cache_dit/quantize/quantize_interface.py,sha256=NG4WP7s8CLW6KhVFb9e1aAjW30KWTCcM2aS5n8QuwsA,1241
+cache_dit/quantize/quantize_svdq.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+cache_dit-0.2.25.dist-info/licenses/LICENSE,sha256=Dqb07Ik2dV41s9nIdMUbiRWEfDqo7-dQeRiY7kPO8PE,3769
+cache_dit-0.2.25.dist-info/METADATA,sha256=a5wbENMZ9BDjHbM3Ejb7Il7x4QzD8W7Lzmu4poo95Wo,19913
+cache_dit-0.2.25.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+cache_dit-0.2.25.dist-info/entry_points.txt,sha256=FX2gysXaZx6NeK1iCLMcIdP8Q4_qikkIHtEmi3oWn8o,65
+cache_dit-0.2.25.dist-info/top_level.txt,sha256=ZJDydonLEhujzz0FOkVbO-BqfzO9d_VqRHmZU-3MOZo,10
+cache_dit-0.2.25.dist-info/RECORD,,