compressed-tensors 0.10.3a20250709__py3-none-any.whl → 0.10.3a20250711__py3-none-any.whl

compressed_tensors/compressors/model_compressors/model_compressor.py

@@ -392,8 +392,8 @@ class ModelCompressor:
         for prefix, module in tqdm(model.named_modules(), desc="Compressing model"):
 
             if prefix in module_to_scheme or prefix in sparse_compression_targets:
-                module_device = get_execution_device(module).type
-                is_meta = (module_device == "meta")
+                module_device = get_execution_device(module)
+                is_meta = module_device.type == "meta"
 
                 exec_device = "meta" if is_meta else "cpu"
                 onloading_device = "meta" if is_meta else module_device
@@ -747,12 +747,16 @@ class ModelCompressor:
 
 def map_module_to_scheme(model: Module) -> Dict[str, QuantizationScheme]:
     """
-    Returns a dictionary which maps quantized module names to their quantization schemes
+    Returns a dictionary which maps quantized module names to their quantization
+    schemes. Only includes modules with weight quantization
     """
     return {
         fix_fsdp_module_name(name): module.quantization_scheme
         for name, module in model.named_modules()
-        if is_module_quantized(module)
+        if (
+            hasattr(module, "quantization_scheme") and
+            module.quantization_scheme.weights is not None
+        )
     }
 
 

compressed_tensors/compressors/sparse_compressors/sparse_24_bitmask.py

@@ -178,9 +178,13 @@ def sparse24_bitmask_compress(
 
     if tensor.is_meta:
         num_rows, num_cols = tensor.shape
-        compressed_values = torch.empty((num_rows, num_cols // 2), dtype=tensor.dtype, device="meta")
+        compressed_values = torch.empty(
+            (num_rows, num_cols // 2), dtype=tensor.dtype, device="meta"
+        )
         packed_cols = (num_cols + 7) // 8
-        bitmasks_packed = torch.empty((num_rows, packed_cols), dtype=torch.uint8, device="meta")
+        bitmasks_packed = torch.empty(
+            (num_rows, packed_cols), dtype=torch.uint8, device="meta"
+        )
         return compressed_values, bitmasks_packed
 
     bytemasks = get_24_bytemasks(tensor=tensor)

compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.10.3.a20250709'
+__version__ = version = '0.10.3.a20250711'
 __version_tuple__ = version_tuple = (0, 10, 3)

{compressed_tensors-0.10.3a20250709.dist-info → compressed_tensors-0.10.3a20250711.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250709
+Version: 0.10.3a20250711
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250709.dist-info → compressed_tensors-0.10.3a20250711.dist-info}/RECORD

@@ -1,11 +1,11 @@
 compressed_tensors/__init__.py,sha256=UtKmifNeBCSE2TZSAfduVNNzHY-3V7bLjZ7n7RuXLOE,812
 compressed_tensors/base.py,sha256=73HYH7HY7O2roC89yG_piPFnZwrBfn_i7HmKl90SKc0,875
-compressed_tensors/version.py,sha256=W6EPpFFl6jLcV1lto1KzNfA-q89i7t3iPuaMgP0ddkI,523
+compressed_tensors/version.py,sha256=S11uf7PlZb4aZTrdFIT3CUNVegXSY80RCGN3qLFo2UQ,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=TAuCA55UC_qEl4itA7m4MQTw1wEd9XWgNuXH6vDwb1Q,32908
+compressed_tensors/compressors/model_compressors/model_compressor.py,sha256=MIMgjLK0VJu4bF_CwyvSftFrC5tvIIMfRt6XB0yLjOE,33056
 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
@@ -14,7 +14,7 @@ compressed_tensors/compressors/quantized_compressors/pack_quantized.py,sha256=47
 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/dense.py,sha256=-OujJ1e0iXBvxYVULrIGvAZ9l-IC0mXczZRnimQdgo4,2314
-compressed_tensors/compressors/sparse_compressors/sparse_24_bitmask.py,sha256=4MyERbYOnBgXPr9eCDqwkbAecp4npRBeXNbfdR3H0E0,9218
+compressed_tensors/compressors/sparse_compressors/sparse_24_bitmask.py,sha256=p8cNV-W4TZKaWDlCwjptQyaIrwqlHizZ1Pn4Vx3-ANk,9262
 compressed_tensors/compressors/sparse_compressors/sparse_bitmask.py,sha256=S8vW0FI9ep_XtUQOxj0P5utJt3vKEYOHjWEPp-Xd9aY,5820
 compressed_tensors/compressors/sparse_quantized_compressors/__init__.py,sha256=4f_cwcKXB1nVVMoiKgTFAc8jAPjPLElo-Df_EDm1_xw,675
 compressed_tensors/compressors/sparse_quantized_compressors/marlin_24.py,sha256=7F9J6wgkecitK5hHuqjetZ18HExHIF4QIw1wgm2Y6U8,10099
@@ -61,8 +61,8 @@ 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=DMfZBuUbA6qp_BG_zIWT3ckiEE33K9ob34s-OgzReO4,12057
 compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
-compressed_tensors-0.10.3a20250709.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors-0.10.3a20250709.dist-info/METADATA,sha256=OlbI5O9HtKj34JBHiFL5EKvk4-NnH8O-S97Jg3lrmZk,7031
-compressed_tensors-0.10.3a20250709.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-compressed_tensors-0.10.3a20250709.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors-0.10.3a20250709.dist-info/RECORD,,
+compressed_tensors-0.10.3a20250711.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors-0.10.3a20250711.dist-info/METADATA,sha256=hKMAAe8C0-WWW8kZ40-9a3tEJy0gzknT3u9xQmww6w8,7031
+compressed_tensors-0.10.3a20250711.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+compressed_tensors-0.10.3a20250711.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors-0.10.3a20250711.dist-info/RECORD,,