compressed-tensors 0.10.3a20250814__py3-none-any.whl → 0.11.0__py3-none-any.whl

compressed_tensors/quantization/quant_scheme.py

@@ -57,6 +57,7 @@ class QuantizationScheme(BaseModel):
         inputs = model.input_activations
         outputs = model.output_activations
         weights = model.weights
+        format = model.format
 
         if inputs is not None:
             if inputs.actorder is not None:
@@ -66,6 +67,11 @@ class QuantizationScheme(BaseModel):
             if outputs.actorder is not None:
                 raise ValueError("Cannot apply actorder to output activations")
 
+        if format == CompressionFormat.mixed_precision.value:
+            raise ValueError(
+                "mixed-precision cannot be set as a format for a QuantizationScheme"
+            )
+
         if (
             inputs
             and weights

compressed_tensors/utils/match.py

@@ -15,7 +15,7 @@
 import logging
 import re
 from collections.abc import Generator
-from typing import Iterable, Mapping, Optional, Tuple
+from typing import Iterable, List, Mapping, Optional, Tuple, Union
 
 import torch
 from compressed_tensors.utils.internal import InternalModule
@@ -57,10 +57,10 @@ def match_named_modules(
     unmatched_targets = set(targets)
     for name, module in model.named_modules():
         for target in targets:
-            if is_match(name, module, target, fused):
+            if is_match(name, module, target, fused=fused):
                 unmatched_targets -= {target}
 
-                if not any(is_match(name, module, ign, fused) for ign in ignore):
+                if not is_match(name, module, ignore, fused=fused):
                     yield name, module
 
     if warn_on_fail:
@@ -155,9 +155,7 @@ def match_modules_set(
     for name, module in model.named_modules():
         # match until we get a full set
         for target in targets:
-            if is_match(name, module, target) and not any(
-                is_match(name, module, ign) for ign in ignore
-            ):
+            if is_match(name, module, target, ignore):
                 if matches[target] is not None:
                     raise ValueError(f"Matched a {target} twice before completing set")
                 matches[target] = module
@@ -176,7 +174,8 @@ def match_modules_set(
 def is_match(
     name: str,
     module: torch.nn.Module,
-    target: str,
+    targets: Union[str, Iterable[str]],
+    ignore: Union[str, Iterable[str]] = tuple(),
     fused: Optional[FusedMappping] = None,
 ) -> bool:
     """
@@ -198,8 +197,17 @@ def is_match(
     :fused: optional mapping from suffixes of fused modules to the suffixes of their
         corresponding shards
     """
+    targets = [targets] if isinstance(targets, str) else targets
+    ignore = [ignore] if isinstance(ignore, str) else ignore
+
     return not isinstance(module, InternalModule) and (
-        _match_name(name, target, fused) or _match_class(module, target)
+        any(
+            _match_name(name, target, fused) or _match_class(module, target)
+            for target in targets
+        )
+        and not any(
+            _match_name(name, ign, fused) or _match_class(module, ign) for ign in ignore
+        )
     )
 
 

compressed_tensors/utils/offload.py

@@ -579,6 +579,7 @@ def remove_dispatch(module: torch.nn.Module) -> torch.nn.Module:
     remove_hook_from_module(module, recurse=True)
     if hasattr(module, "hf_device_map"):
         delattr(module, "hf_device_map")
+    module.to("cpu")
 
     return module
 

compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.10.3.a20250814'
-__version_tuple__ = version_tuple = (0, 10, 3)
+__version__ = version = '0.11.0'
+__version_tuple__ = version_tuple = (0, 11, 0)

{compressed_tensors-0.10.3a20250814.dist-info → compressed_tensors-0.11.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250814
+Version: 0.11.0
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250814.dist-info → compressed_tensors-0.11.0.dist-info}/RECORD

@@ -1,6 +1,6 @@
 compressed_tensors/__init__.py,sha256=UtKmifNeBCSE2TZSAfduVNNzHY-3V7bLjZ7n7RuXLOE,812
 compressed_tensors/base.py,sha256=-gxWvDF4LCkyeDP8YlGzvBBKxo4Dk9h4NINPD61drFU,921
-compressed_tensors/version.py,sha256=fAUC53w9XJ-gbZ3V6UhPrss4y7OIGxsBJ0rFa1T--zA,523
+compressed_tensors/version.py,sha256=h33CazvO6X2vAgSb_I-BIOLNm7W5MVFnfSYRgLiRpMA,513
 compressed_tensors/compressors/__init__.py,sha256=smSygTSfcfuujRrAXDc6uZm4L_ccV1tWZewqVnOb4lM,825
 compressed_tensors/compressors/base.py,sha256=nvWsv4xEw1Tkxkxth6TmHplDYXfBeP22xWxOsZERyDY,7204
 compressed_tensors/compressors/helpers.py,sha256=OK6qxX9j3bHwF9JfIYSGMgBJe2PWjlTA3byXKCJaTIQ,5431
@@ -28,7 +28,7 @@ compressed_tensors/linear/compressed_linear.py,sha256=1yo9RyjA0aQ--iuIknFfcSorJn
 compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
 compressed_tensors/quantization/quant_args.py,sha256=PMoaa6hpyJLGGSeCWefGmzGVxbOtxAdDunHJi_L5gNs,12894
 compressed_tensors/quantization/quant_config.py,sha256=2NgDwKuQn0f-ojiHC8c6tXtYX_zQlk26Rj-bU71QKvA,10598
-compressed_tensors/quantization/quant_scheme.py,sha256=k25Cdx7BZCvLlRlENu4BVoFxquqcErP58P3Y_1HsKB4,8661
+compressed_tensors/quantization/quant_scheme.py,sha256=aTdGWz_JenLlTqCeNFcujCJjgWQYkXEq2FXoN0YSNqI,8883
 compressed_tensors/quantization/lifecycle/__init__.py,sha256=_uItzFWusyV74Zco_pHLOTdE9a83cL-R-ZdyQrBkIyw,772
 compressed_tensors/quantization/lifecycle/apply.py,sha256=wM8mVcbKvZjBo18pSXMp28i30YWwUXJPSS7_HCakH9U,17892
 compressed_tensors/quantization/lifecycle/compressed.py,sha256=Fj9n66IN0EWsOAkBHg3O0GlOQpxstqjCcs0ttzMXrJ0,2296
@@ -56,15 +56,15 @@ compressed_tensors/transform/utils/matrix.py,sha256=FIHCUlpWVIIhdr3c6EbQec41JeiP
 compressed_tensors/utils/__init__.py,sha256=spzbjUO4-hZ2jXGST27r3MIt2yzIXsjdbEaYyaMcizo,873
 compressed_tensors/utils/helpers.py,sha256=Q3iRAa2XSdmmn4vSpUplnvKOmWwn4Clao9ZkPBHXtpI,12604
 compressed_tensors/utils/internal.py,sha256=7SSWgDoNFRnlfadwkoFhLW-T2jOc7Po_WzWv5h32Sa8,982
-compressed_tensors/utils/match.py,sha256=9x-yZIlq7ndSLf2aQwNT7IpBQDe-8H6utiJkji8wPrQ,9397
-compressed_tensors/utils/offload.py,sha256=gFoEDaissHsLM5-JDbgPxh5hiE9VFN4HFxvszYvReos,24446
+compressed_tensors/utils/match.py,sha256=wXzcfUWlVE5Wc_OAvN7oUFqLhVHJ83oSt1Nhzw01igg,9701
+compressed_tensors/utils/offload.py,sha256=7EDCtRbnw5gZQlZxTAICrc7ONeNW0pKdQFqOeqGlF6g,24467
 compressed_tensors/utils/permutations_24.py,sha256=kx6fsfDHebx94zsSzhXGyCyuC9sVyah6BUUir_StT28,2530
 compressed_tensors/utils/permute.py,sha256=V6tJLKo3Syccj-viv4F7ZKZgJeCB-hl-dK8RKI_kBwI,2355
 compressed_tensors/utils/safetensors_load.py,sha256=DMfZBuUbA6qp_BG_zIWT3ckiEE33K9ob34s-OgzReO4,12057
 compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
 compressed_tensors/utils/type.py,sha256=bNwoo_FWlvLuDpYAGGzZJITRg0JA_Ngk9LGPo-kvjeU,2554
-compressed_tensors-0.10.3a20250814.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors-0.10.3a20250814.dist-info/METADATA,sha256=-7voWXyJPB13WkMJADa57hDE4euKxrCjnQfYdHROjKg,7031
-compressed_tensors-0.10.3a20250814.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-compressed_tensors-0.10.3a20250814.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors-0.10.3a20250814.dist-info/RECORD,,
+compressed_tensors-0.11.0.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors-0.11.0.dist-info/METADATA,sha256=RBt5uf5--bJvf_NPxjaOCX0HoHgHVUN9a8fmaIyXCAM,7022
+compressed_tensors-0.11.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+compressed_tensors-0.11.0.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors-0.11.0.dist-info/RECORD,,