compressed-tensors-nightly 0.9.2.20250309__py3-none-any.whl → 0.9.2.20250312__py3-none-any.whl

compressed_tensors/linear/compressed_linear.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import warnings
 from typing import Dict, Tuple
 
 import torch
@@ -33,14 +34,15 @@ class CompressedLinear(Linear):
     Wrapper module for running a compressed forward pass of a quantized Linear module.
     The wrapped layer will decompressed on each forward call.
 
-    :param module: dense linear module to replace
-    :param quantization_scheme: quantization config for the module to wrap
-    :param quantization_format: compression format module is stored as
     """
 
     def __init__(self, *args, **kwargs) -> None:
         super().__init__(*args, **kwargs)
-        self._is_compressed = True
+        warnings.warn(
+            "CompressedLinear should not be initialized directly. "
+            "Use the from_linear method instead.",
+            UserWarning,
+        )
 
     @classmethod
     @torch.no_grad()
@@ -50,6 +52,12 @@ class CompressedLinear(Linear):
         quantization_scheme: QuantizationScheme,
         quantization_format: str,
     ):
+        """
+        :param module: dense linear module to replace
+        :param quantization_scheme: quantization config for the module to wrap
+        :param quantization_format: compression format module is stored as
+        :return: CompressedLinear module wrapping the input module
+        """
         module.__class__ = CompressedLinear
         module.compressor = BaseCompressor.load_from_registry(quantization_format)
         device = next(module.parameters()).device
@@ -90,8 +98,9 @@ class CompressedLinear(Linear):
         """
         Decompresses the weight, then runs the wrapped forward pass
         """
-        if self._is_compressed:
-            self.weight = self.compressor.decompress_module(self)
-            self._is_compressed = False
+        if self.quantization_status == QuantizationStatus.COMPRESSED:
+            decompressed_weight = self.compressor.decompress_module(self)
+            self.weight.data = decompressed_weight
+            self.quantization_status = QuantizationStatus.FROZEN
 
         return linear(input, self.weight, self.bias)

{compressed_tensors_nightly-0.9.2.20250309.dist-info → compressed_tensors_nightly-0.9.2.20250312.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: compressed-tensors-nightly
-Version: 0.9.2.20250309
+Version: 0.9.2.20250312
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors_nightly-0.9.2.20250309.dist-info → compressed_tensors_nightly-0.9.2.20250312.dist-info}/RECORD

@@ -23,7 +23,7 @@ compressed_tensors/config/dense.py,sha256=NgSxnFCnckU9-iunxEaqiFwqgdO7YYxlWKR74j
 compressed_tensors/config/sparse_24_bitmask.py,sha256=Lhj39zT2V1hxftprvxvneyhv45ShlXOKd75DBbDTyTE,1401
 compressed_tensors/config/sparse_bitmask.py,sha256=pZUboRNZTu6NajGOQEFExoPknak5ynVAUeiiYpS1Gt8,1308
 compressed_tensors/linear/__init__.py,sha256=fH6rjBYAxuwrTzBTlTjTgCYNyh6TCvCqajCz4Im4YrA,617
-compressed_tensors/linear/compressed_linear.py,sha256=TnEVJCwip4L21YNayon9ndpFU9dinrNKPz1z1w2U2NE,3529
+compressed_tensors/linear/compressed_linear.py,sha256=4553oRbnuMW1siVtnHk7Dg617iVQzu2kRSX5WyLZzxk,3912
 compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
 compressed_tensors/quantization/quant_args.py,sha256=sKpb8DcNObidjXjNol1Tn_Iih3ZXBycSp-fyz68TGhY,9117
 compressed_tensors/quantization/quant_config.py,sha256=vx06wBo91p4LCb3Vzd-2eCTUeIf_Sz2ZXRP263eQyjQ,10385
@@ -45,8 +45,8 @@ compressed_tensors/utils/permutations_24.py,sha256=kx6fsfDHebx94zsSzhXGyCyuC9sVy
 compressed_tensors/utils/permute.py,sha256=V6tJLKo3Syccj-viv4F7ZKZgJeCB-hl-dK8RKI_kBwI,2355
 compressed_tensors/utils/safetensors_load.py,sha256=5SeM2hzLh77Ne8Vk7qR6-km7cf8bhov41ExpWITqX3A,11470
 compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
-compressed_tensors_nightly-0.9.2.20250309.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors_nightly-0.9.2.20250309.dist-info/METADATA,sha256=fD_KdgGLHD18bXERgNchyMtqDk-hhtmfc1Dr9nG8mnw,6992
-compressed_tensors_nightly-0.9.2.20250309.dist-info/WHEEL,sha256=jB7zZ3N9hIM9adW7qlTAyycLYW9npaWKLRzaoVcLKcM,91
-compressed_tensors_nightly-0.9.2.20250309.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors_nightly-0.9.2.20250309.dist-info/RECORD,,
+compressed_tensors_nightly-0.9.2.20250312.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors_nightly-0.9.2.20250312.dist-info/METADATA,sha256=OsZMbP2kxO9dpksHaqRdcNSe6HLGXmdqyqb9bGuRZtw,6992
+compressed_tensors_nightly-0.9.2.20250312.dist-info/WHEEL,sha256=52BFRY2Up02UkjOa29eZOS2VxUrpPORXg1pkohGGUS8,91
+compressed_tensors_nightly-0.9.2.20250312.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors_nightly-0.9.2.20250312.dist-info/RECORD,,

{compressed_tensors_nightly-0.9.2.20250309.dist-info → compressed_tensors_nightly-0.9.2.20250312.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (75.8.2)
+Generator: setuptools (76.0.0)
 Root-Is-Purelib: true
 Tag: py3-none-any