compressed-tensors 0.10.3a20250811__py3-none-any.whl → 0.10.3a20250812__py3-none-any.whl

compressed_tensors/compressors/model_compressors/model_compressor.py

@@ -207,7 +207,7 @@ class ModelCompressor:
 
     @staticmethod
     def parse_sparsity_config(
-        compression_config: Union[Dict[str, Any], "CompressedTensorsConfig"]
+        compression_config: Union[Dict[str, Any], "CompressedTensorsConfig"],
     ) -> Union[Dict[str, Any], None]:
         """
         Parse sparsity config from quantization/compression config. Sparsity
@@ -227,7 +227,7 @@ class ModelCompressor:
 
     @staticmethod
     def parse_quantization_config(
-        compression_config: Union[Dict[str, Any], "CompressedTensorsConfig"]
+        compression_config: Union[Dict[str, Any], "CompressedTensorsConfig"],
     ) -> Union[Dict[str, Any], None]:
         """
         Parse quantization config from quantization/compression config. The
@@ -246,6 +246,7 @@ class ModelCompressor:
 
         quantization_config = deepcopy(compression_config)
         quantization_config.pop(SPARSITY_CONFIG_NAME, None)
+        quantization_config.pop(TRANSFORM_CONFIG_NAME, None)
 
         # some fields are required, even if a qconfig is not present
         # pop them off and if nothing remains, then there is no qconfig
@@ -674,7 +675,7 @@ class ModelCompressor:
 
         # serialize configs into json
         qconfig_data = (
-            self.quantization_config.model_dump(exclude=["quant_method", "format"])
+            self.quantization_config.model_dump(exclude=["quant_method"])
             if self.quantization_config is not None
             else {}
         )

compressed_tensors/utils/offload.py

@@ -86,6 +86,7 @@ __all__ = [
     "offloaded_dispatch",
     "disable_offloading",
     "remove_dispatch",
+    "cast_to_device",
 ]
 
 
@@ -169,6 +170,19 @@ def update_parameter_data(
 """ Candidates for Upstreaming """
 
 
+def cast_to_device(device_spec: Union[int, torch.device]) -> torch.device:
+    """
+    Convert an integer device index or torch.device into a torch.device object.
+
+    :param device_spec: Device index (int) or torch.device object.
+                        Negative integers map to CPU.
+    :return: torch.device corresponding to the given device specification.
+    """
+    if isinstance(device_spec, int):
+        return torch.device(f"cuda:{device_spec}" if device_spec >= 0 else "cpu")
+    return device_spec
+
+
 def get_execution_device(module: torch.nn.Module) -> torch.device:
     """
     Get the device which inputs should be moved to before module execution.
@@ -179,7 +193,7 @@ def get_execution_device(module: torch.nn.Module) -> torch.device:
     """
     for submodule in module.modules():
         if has_offloaded_params(submodule):
-            return submodule._hf_hook.execution_device
+            return cast_to_device(submodule._hf_hook.execution_device)
 
         param = next(submodule.parameters(recurse=False), None)
         if param is not None:

compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.10.3.a20250811'
+__version__ = version = '0.10.3.a20250812'
 __version_tuple__ = version_tuple = (0, 10, 3)

{compressed_tensors-0.10.3a20250811.dist-info → compressed_tensors-0.10.3a20250812.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250811
+Version: 0.10.3a20250812
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250811.dist-info → compressed_tensors-0.10.3a20250812.dist-info}/RECORD

@@ -1,11 +1,11 @@
 compressed_tensors/__init__.py,sha256=UtKmifNeBCSE2TZSAfduVNNzHY-3V7bLjZ7n7RuXLOE,812
 compressed_tensors/base.py,sha256=-gxWvDF4LCkyeDP8YlGzvBBKxo4Dk9h4NINPD61drFU,921
-compressed_tensors/version.py,sha256=9NgEdMzgL7r039RghUuU-BhjHVuGK1utk2z8Au9OlWA,523
+compressed_tensors/version.py,sha256=xvYY_n0Nd8St_3mnUNQJLLacYomaXp2WG5TBIJ3wSDo,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=0WULLKpgWuTQLjKsCstiTssT778wp9TWMkQjHbYO4Zo,33989
+compressed_tensors/compressors/model_compressors/model_compressor.py,sha256=NEnRjTRhmOHP_TVgrhQb58t-XjftDNoZBNyTXQQBRao,34042
 compressed_tensors/compressors/quantized_compressors/__init__.py,sha256=KvaFBL_Q84LxRGJOV035M8OBoCkAx8kOkfphswgkKWk,745
 compressed_tensors/compressors/quantized_compressors/base.py,sha256=YGUMzbxekj_36ChgQnVZN6T8uDjXtGG1zfMIBGBLWco,10354
 compressed_tensors/compressors/quantized_compressors/naive_quantized.py,sha256=0ANDcuD8aXPqTYNPY6GnX9iS6eXJw6P0TzNV_rYS2l8,5369
@@ -57,14 +57,14 @@ compressed_tensors/utils/__init__.py,sha256=spzbjUO4-hZ2jXGST27r3MIt2yzIXsjdbEaY
 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=3XiBuWbUkBAt8v1t5i57qDcbB3VJQs_FDeayi-JzIWg,23896
+compressed_tensors/utils/offload.py,sha256=gFoEDaissHsLM5-JDbgPxh5hiE9VFN4HFxvszYvReos,24446
 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-0.10.3a20250811.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors-0.10.3a20250811.dist-info/METADATA,sha256=mYMXLEK9r53lXrMbZBRmkimI3aW-X1x4n-8DUThb0K8,7031
-compressed_tensors-0.10.3a20250811.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-compressed_tensors-0.10.3a20250811.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors-0.10.3a20250811.dist-info/RECORD,,
+compressed_tensors-0.10.3a20250812.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors-0.10.3a20250812.dist-info/METADATA,sha256=FKQ6iJRJ_Q7GMoCLI73bfQXEKxW5E2hgmDl9lCTbtSY,7031
+compressed_tensors-0.10.3a20250812.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+compressed_tensors-0.10.3a20250812.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors-0.10.3a20250812.dist-info/RECORD,,