compressed-tensors 0.10.2a20250606__py3-none-any.whl → 0.10.2a20250609__py3-none-any.whl

compressed_tensors/compressors/model_compressors/model_compressor.py

@@ -68,6 +68,10 @@ from transformers import AutoConfig
 from transformers.file_utils import CONFIG_NAME
 
 
+if TYPE_CHECKING:
+    from compressed_tensors.compressors import BaseQuantizationCompressor
+
+
 __all__ = ["ModelCompressor", "map_module_to_scheme"]
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
@@ -257,7 +261,9 @@ class ModelCompressor:
         self.sparsity_config = sparsity_config
         self.quantization_config = quantization_config
         self.sparsity_compressor = None
-        self.quantization_compressor = None
+        self.quantization_compressor: Optional[
+            Union[BaseQuantizationCompressor, DenseCompressor]
+        ] = None
 
         if sparsity_config is not None:
             self.sparsity_compressor = BaseCompressor.load_from_registry(

compressed_tensors/compressors/sparse_compressors/base.py

@@ -13,8 +13,9 @@
 # limitations under the License.
 
 import logging
-from typing import Dict, Generator, Optional, Set, Tuple
+from typing import TYPE_CHECKING, Dict, Generator, Optional, Set, Tuple
 
+import torch
 from compressed_tensors.compressors.base import BaseCompressor
 from compressed_tensors.utils import (
     get_nested_mappings_from_state_dict,
@@ -26,6 +27,10 @@ from torch import Tensor
 from tqdm import tqdm
 
 
+if TYPE_CHECKING:
+    from compressed_tensors.quantization import QuantizationScheme
+
+
 __all__ = ["BaseSparseCompressor"]
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
@@ -200,3 +205,16 @@ class BaseSparseCompressor(BaseCompressor):
         return (
             name.endswith(".weight") and name[: -(len(".weight"))] in expanded_targets
         )
+
+    def decompress_module_from_state_dict(
+        self,
+        prefix: str,
+        state_dict: Dict[str, torch.Tensor],
+        scheme: "QuantizationScheme",
+    ) -> Dict[str, torch.Tensor]:
+        """
+        This function is implemented as a workaround because of how
+        `ModelCompressor.quantization_compressor` can be set to either
+        an instance of `BaseQuantizationCompressor` or `BaseSparseCompressor`.
+        """
+        return state_dict.copy()

compressed_tensors/utils/offload.py

@@ -87,12 +87,15 @@ def check_accelerate(fallback: Any):
         if not _has_accelerate:
 
             if fallback == "error":
+
                 @wraps(func)
                 def fallback_fn(*args, **kwargs):
                     raise ValueError(
                         "Please install `accelerate` in order to use this function"
                     )
+
             else:
+
                 @wraps(func)
                 def fallback_fn(*args, **kwargs):
                     return fallback

compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.10.2.a20250606'
+__version__ = version = '0.10.2.a20250609'
 __version_tuple__ = version_tuple = (0, 10, 2)

{compressed_tensors-0.10.2a20250606.dist-info → compressed_tensors-0.10.2a20250609.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.2a20250606
+Version: 0.10.2a20250609
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.2a20250606.dist-info → compressed_tensors-0.10.2a20250609.dist-info}/RECORD

@@ -1,18 +1,18 @@
 compressed_tensors/__init__.py,sha256=UtKmifNeBCSE2TZSAfduVNNzHY-3V7bLjZ7n7RuXLOE,812
 compressed_tensors/base.py,sha256=73HYH7HY7O2roC89yG_piPFnZwrBfn_i7HmKl90SKc0,875
-compressed_tensors/version.py,sha256=g4m3RqR1QAndk8dMBzUoODghgITncD4Bp_Td-dPvrbQ,523
+compressed_tensors/version.py,sha256=LTla_Pj0LNUtkAT7ETF33M0rqtov9c8Q6T5tjzwvHOs,523
 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
 compressed_tensors/compressors/model_compressors/__init__.py,sha256=5RGGPFu4YqEt_aOdFSQYFYFDjcZFJN0CsMqRtDZz3Js,666
-compressed_tensors/compressors/model_compressors/model_compressor.py,sha256=72h2tWDIGbbqLQF8MDzOehy18eu5TvsCLd_AuzGv_O4,32517
+compressed_tensors/compressors/model_compressors/model_compressor.py,sha256=nt0KxhZakDdlTIebBYcSvqxLCZhA6p6IL_1AYiHLFug,32695
 compressed_tensors/compressors/quantized_compressors/__init__.py,sha256=KvaFBL_Q84LxRGJOV035M8OBoCkAx8kOkfphswgkKWk,745
 compressed_tensors/compressors/quantized_compressors/base.py,sha256=ByE3z61boZ5wdz0nhc-2CJH61bSixJQE78pfkS6XRDg,10269
 compressed_tensors/compressors/quantized_compressors/naive_quantized.py,sha256=0ANDcuD8aXPqTYNPY6GnX9iS6eXJw6P0TzNV_rYS2l8,5369
 compressed_tensors/compressors/quantized_compressors/nvfp4_quantized.py,sha256=Gw-lVzk5jrKUlM5UTCiJBmhM5gHzB9mn8r298MVUbDI,6395
 compressed_tensors/compressors/quantized_compressors/pack_quantized.py,sha256=_66tQ8bxslDUdas-ULORXblPw9kdNNn1UJJU9-ZOGPY,11380
 compressed_tensors/compressors/sparse_compressors/__init__.py,sha256=Atuz-OdEgn8OCUhx7Ovd6gXdyImAI186uCR-uR0t_Nk,737
-compressed_tensors/compressors/sparse_compressors/base.py,sha256=YNZWcHjDleAlqbgRZQ6oJf44MQb_UDNvJGOqhl26uFA,8098
+compressed_tensors/compressors/sparse_compressors/base.py,sha256=4-7Hm2uJ4qYtl4hi6GqwegGg2gYNhbq7noledK3h86Q,8684
 compressed_tensors/compressors/sparse_compressors/dense.py,sha256=rPaxbP7P52prWNs4lGaiBbpNvsQLElFMwOrq1oBP2Yg,1733
 compressed_tensors/compressors/sparse_compressors/sparse_24_bitmask.py,sha256=4cwkj40SFrXEyE_jyt2xjz3R-gTdU9uMpMFUKo1pRBA,8643
 compressed_tensors/compressors/sparse_compressors/sparse_bitmask.py,sha256=S8vW0FI9ep_XtUQOxj0P5utJt3vKEYOHjWEPp-Xd9aY,5820
@@ -48,13 +48,13 @@ compressed_tensors/transform/utils/hadamard.py,sha256=SmPZmnHtc5N36gJA5EbM1T65uf
 compressed_tensors/transform/utils/utils.py,sha256=PRPTYwPs2nnNaQMq2GEbC4QYKHFKlZwaRyPgdDhl66g,2992
 compressed_tensors/utils/__init__.py,sha256=gS4gSU2pwcAbsKj-6YMaqhm25udFy6ISYaWBf-myRSM,808
 compressed_tensors/utils/helpers.py,sha256=cPg-ikdeA92aIGwBONg8GmPNvcGlFhozyJVwsRiXBTA,11981
-compressed_tensors/utils/offload.py,sha256=fT7WiUQmRmJ2Reb3I5kNcsHy4YdmZJHSOTNdS0tbKQo,20316
+compressed_tensors/utils/offload.py,sha256=aGcBv0UWLuauY_-HBSei5Gm0xsZOfWn7NeX76tcLf30,20319
 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-0.10.2a20250606.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors-0.10.2a20250606.dist-info/METADATA,sha256=zQOy95JRCub4it-SD7sRzzV_UCQoMOgQls9jFlMKk8o,7005
-compressed_tensors-0.10.2a20250606.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-compressed_tensors-0.10.2a20250606.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors-0.10.2a20250606.dist-info/RECORD,,
+compressed_tensors-0.10.2a20250609.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors-0.10.2a20250609.dist-info/METADATA,sha256=_OH8bysJ6ymrDyfkHXrKuZNb47qh38hfdCCOt7RRYeM,7005
+compressed_tensors-0.10.2a20250609.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+compressed_tensors-0.10.2a20250609.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors-0.10.2a20250609.dist-info/RECORD,,