compressed-tensors 0.11.1a20250908__py3-none-any.whl → 0.11.1a20250909__py3-none-any.whl

compressed_tensors/compressors/quantized_compressors/nvfp4_quantized.py

@@ -140,6 +140,11 @@ def pack_fp4_to_uint8(x: torch.Tensor) -> torch.Tensor:
     m, n = x.shape
     device = x.device
 
+    if n % 2 != 0:
+        raise ValueError(
+            "tensor must have an even number of columns for nvfp4 compression"
+        )
+
     # Create lookup table for FP4 values to indices
     # Map the absolute values to 0-7 indices
     kE2M1 = torch.tensor(FLOAT_TO_E2M1, device=device, dtype=x.dtype)
@@ -155,10 +160,6 @@ def pack_fp4_to_uint8(x: torch.Tensor) -> torch.Tensor:
     # Reshape to prepare for packing pairs of values
     indices = indices.reshape(-1)
 
-    # Handle odd length by padding if necessary
-    if indices.numel() % 2 != 0:
-        indices = torch.cat([indices, torch.zeros(1, dtype=torch.long, device=device)])
-
     # Reshape to pair consecutive elements
     indices = indices.reshape(-1, 2)
 

compressed_tensors/quantization/quant_scheme.py

@@ -63,9 +63,19 @@ class QuantizationScheme(BaseModel):
             if inputs.strategy not in (
                 QuantizationStrategy.TOKEN,
                 QuantizationStrategy.TENSOR,
+                QuantizationStrategy.GROUP,
                 QuantizationStrategy.TENSOR_GROUP,
             ):
-                raise ValueError(
+                if (
+                    inputs.strategy == QuantizationStrategy.GROUP
+                    and inputs.dynamic is True
+                ):
+                    raise NotImplementedError(
+                        "Static and local group-wise activation "
+                        "quantization is not supported"
+                    )
+
+                raise NotImplementedError(
                     f"Using {inputs.strategy} strategy is not supported for "
                     "activation quantization"
                 )

compressed_tensors/utils/offload.py

@@ -68,10 +68,8 @@ except ImportError:
 
 
 __all__ = [
-    "is_module_offloaded",
     "get_execution_device",
     "get_offloaded_device",
-    "update_prefix_dict",
     "update_parameter_data",
     "register_offload_parameter",
     "update_offload_parameter",
@@ -117,11 +115,6 @@ def check_accelerate(fallback: Any):
 """ Candidates for Depreciation """
 
 
-@check_accelerate(fallback=False)
-def is_module_offloaded(module: torch.nn.Module) -> bool:
-    return has_offloaded_params(module)
-
-
 def get_offloaded_device(module: torch.nn.Module) -> torch.device:
     """
     :param module: module to check
@@ -137,25 +130,6 @@ def get_offloaded_device(module: torch.nn.Module) -> torch.device:
         return get_execution_device(module)
 
 
-@check_accelerate(fallback=None)
-def update_prefix_dict(module: torch.nn.Module, key: str, data: torch.Tensor):
-    """
-    Updates the offloaded state dict for a given module. Parameter named key is replaced
-    by data. This is neccesary because parameter updates for offloaded modules do not
-    persist automatically between loads. This function only affects the offloaded
-    state dict and not the current state of the loaded module.
-
-    :param module: module containing the parameter to update
-    :param key: name of parameter to update
-    :param data: tensor to update parameter with in the offloaded state dict
-    """
-    if not has_offloaded_params(module):
-        raise ValueError("Prefix dict is only applicable to offloaded modules")
-
-    weights_map = module._hf_hook.weights_map
-    offload_to_weights_map(weights_map, key, data)
-
-
 def update_parameter_data(
     module: torch.nn.Module, new_param_data: torch.Tensor, param_name: str
 ):

compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.11.1.a20250908'
+__version__ = version = '0.11.1.a20250909'
 __version_tuple__ = version_tuple = (0, 11, 1)

{compressed_tensors-0.11.1a20250908.dist-info → compressed_tensors-0.11.1a20250909.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.11.1a20250908
+Version: 0.11.1a20250909
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.11.1a20250908.dist-info → compressed_tensors-0.11.1a20250909.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=8qf_B1P1NNbEDyEkRyxNNhdvTofGEV0EE02UMN3na5k,523
+compressed_tensors/version.py,sha256=r7NPIWZc4XFCqdYyi4qPxtVWw1N9RBvLtcldfOSxGIA,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
@@ -9,7 +9,7 @@ compressed_tensors/compressors/model_compressors/model_compressor.py,sha256=mZqp
 compressed_tensors/compressors/quantized_compressors/__init__.py,sha256=KvaFBL_Q84LxRGJOV035M8OBoCkAx8kOkfphswgkKWk,745
 compressed_tensors/compressors/quantized_compressors/base.py,sha256=rWvaWDqzi8cctBo982g2n3-y6afRiFl3jfTd90lSMrY,10413
 compressed_tensors/compressors/quantized_compressors/naive_quantized.py,sha256=0ANDcuD8aXPqTYNPY6GnX9iS6eXJw6P0TzNV_rYS2l8,5369
-compressed_tensors/compressors/quantized_compressors/nvfp4_quantized.py,sha256=Qq790d5VQQccq6Dj8YhBwhr7S3DqMJNoYPI5S6M1FNo,7183
+compressed_tensors/compressors/quantized_compressors/nvfp4_quantized.py,sha256=lrF-FVAZ2OGAd1yFvH5tp7d5_yvi37aZEuJui--7RmQ,7148
 compressed_tensors/compressors/quantized_compressors/pack_quantized.py,sha256=D8h9ltxSIYi1XEKYgbYu1ebbXzCibhPi-eZsBUi0NOg,11245
 compressed_tensors/compressors/sparse_compressors/__init__.py,sha256=Atuz-OdEgn8OCUhx7Ovd6gXdyImAI186uCR-uR0t_Nk,737
 compressed_tensors/compressors/sparse_compressors/base.py,sha256=YNZWcHjDleAlqbgRZQ6oJf44MQb_UDNvJGOqhl26uFA,8098
@@ -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=5AxYKqCSlg7CDgz2N8G4ZRVIiSUKvIm-SCQa-Bq_SF0,12916
 compressed_tensors/quantization/quant_config.py,sha256=2NgDwKuQn0f-ojiHC8c6tXtYX_zQlk26Rj-bU71QKvA,10598
-compressed_tensors/quantization/quant_scheme.py,sha256=2pV3tPNgo6ovi6FLxP4ZFznEmInlC1L90Pq9I1HI_Xk,9275
+compressed_tensors/quantization/quant_scheme.py,sha256=EG86Bq5c8q1O4fJL_o3s7gOu1S5SrcLjfNYOPDn414A,9673
 compressed_tensors/quantization/lifecycle/__init__.py,sha256=_uItzFWusyV74Zco_pHLOTdE9a83cL-R-ZdyQrBkIyw,772
 compressed_tensors/quantization/lifecycle/apply.py,sha256=TuSjKomSk4N0My-UY9PWk2Nyuze6TilEGPsZELgotzk,14716
 compressed_tensors/quantization/lifecycle/compressed.py,sha256=Fj9n66IN0EWsOAkBHg3O0GlOQpxstqjCcs0ttzMXrJ0,2296
@@ -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=y03xJyWTXV8bjIPN5Z4S0_w797qMnh-Z4aiPEGQ4zNE,11239
-compressed_tensors/utils/offload.py,sha256=b0Q2P0hJLQBGEqdRwOh6SOK3_eJCqHNnIX38-wqeef0,24577
+compressed_tensors/utils/offload.py,sha256=eXqLzl8kUkVDlNtcO5sn_4QoDcbAaxbCAS3tyZ-aGr8,23538
 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=Vql34aCTDHwmTZXJHzCyBISJo7iA7EQ78LdTlMjdpZo,12023
 compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
 compressed_tensors/utils/type.py,sha256=bNwoo_FWlvLuDpYAGGzZJITRg0JA_Ngk9LGPo-kvjeU,2554
-compressed_tensors-0.11.1a20250908.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors-0.11.1a20250908.dist-info/METADATA,sha256=lLokpvLLt0OOrfW9axfgwLmBhqCR7IqyjLCQFRdxCsU,7031
-compressed_tensors-0.11.1a20250908.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-compressed_tensors-0.11.1a20250908.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors-0.11.1a20250908.dist-info/RECORD,,
+compressed_tensors-0.11.1a20250909.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors-0.11.1a20250909.dist-info/METADATA,sha256=ZESMNkRHo7FRcoSr9v_JKDGml5oWDK-Tgcboj-0CnE4,7031
+compressed_tensors-0.11.1a20250909.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+compressed_tensors-0.11.1a20250909.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors-0.11.1a20250909.dist-info/RECORD,,