compressed-tensors-nightly 0.5.0.20240804__py3-none-any.whl → 0.5.0.20240806__py3-none-any.whl

compressed_tensors/quantization/lifecycle/apply.py

@@ -15,6 +15,7 @@
 import logging
 import re
 from collections import OrderedDict
+from copy import deepcopy
 from typing import Dict, Iterable, List, Optional
 from typing import OrderedDict as OrderedDictType
 from typing import Union
@@ -110,6 +111,10 @@ def apply_quantization_config(model: Module, config: QuantizationConfig) -> Dict
     :param model: model to apply quantization config to
     :param config: quantization config
     """
+    # remove reference to the original `config`
+    # argument. This function can mutate it, and we'd
+    # like to keep the original `config` as it is.
+    config = deepcopy(config)
     # build mapping of targets to schemes for easier matching
     # use ordered dict to preserve target ordering in config
     target_to_scheme = OrderedDict()

compressed_tensors/quantization/quant_config.py

@@ -103,7 +103,7 @@ LIFECYCLE_ORDER = [
     QuantizationStatus.COMPRESSED,
 ]
 
-DEFAULT_QUANTIZATION_METHOD = "compressed-tensors"
+DEFAULT_QUANTIZATION_METHOD = "compressed_tensors"
 DEFAULT_QUANTIZATION_FORMAT = "fakequant"
 
 

compressed_tensors/utils/offload.py

@@ -40,7 +40,13 @@ def get_execution_device(module: Module) -> torch.device:
     """
     if is_module_offloaded(module):
         return module._hf_hook.execution_device
-    return next(module.parameters()).device
+    device = next(module.parameters()).device
+
+    # offload only gets set for leaf modules, fallback to checking for device type
+    if device.type == "meta":
+        return module._hf_hook.execution_device
+
+    return device
 
 
 def get_offloaded_device(module: Module) -> torch.device:

{compressed_tensors_nightly-0.5.0.20240804.dist-info → compressed_tensors_nightly-0.5.0.20240806.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: compressed-tensors-nightly
-Version: 0.5.0.20240804
+Version: 0.5.0.20240806
 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-0.5.0.20240804.dist-info → compressed_tensors_nightly-0.5.0.20240806.dist-info}/RECORD

@@ -16,10 +16,10 @@ compressed_tensors/config/dense.py,sha256=NgSxnFCnckU9-iunxEaqiFwqgdO7YYxlWKR74j
 compressed_tensors/config/sparse_bitmask.py,sha256=pZUboRNZTu6NajGOQEFExoPknak5ynVAUeiiYpS1Gt8,1308
 compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
 compressed_tensors/quantization/quant_args.py,sha256=Vc_tWSTcbZZsMJlACpLq4JEPvGx87izc8VEx-mcXjoM,5621
-compressed_tensors/quantization/quant_config.py,sha256=NpVu8YJ4Xw2pIQW_PGaNaml8kx1bUnxkvb0jBYWbKdE,9971
+compressed_tensors/quantization/quant_config.py,sha256=a9U2Iydwi0AIZ30-7J403nBBBHXVCNEzz4HbvnD4_B8,9971
 compressed_tensors/quantization/quant_scheme.py,sha256=_RKOFJI0T5xJVBLX63UeYkSY4EFAecsBnqzUIVBjeU0,6014
 compressed_tensors/quantization/lifecycle/__init__.py,sha256=MXE2E7GfIfRRfhrdGy2Og3AZOz5N59B0ZGFcsD89y6c,821
-compressed_tensors/quantization/lifecycle/apply.py,sha256=aamouy1IWCSGl5_lfC7rZrUNbMEfhyHQrZFGEm2VH4w,13242
+compressed_tensors/quantization/lifecycle/apply.py,sha256=FDKw6AaQjOwNtcX6oFXNx5b_bwm6BpuFtly1Ll6-WFE,13451
 compressed_tensors/quantization/lifecycle/calibration.py,sha256=n-m4xwa9Ds2xrltp6r0rhuVLJhE8bQ1LnifrcrbA-ig,2448
 compressed_tensors/quantization/lifecycle/compressed.py,sha256=VreB10xPwgSLQQlTu20UCrFpRS--cA7-lx5s7nrPPrg,2247
 compressed_tensors/quantization/lifecycle/forward.py,sha256=6PSXYcf-R1dOY8zsuIWnBaoyARNymYc3-qvV6-L7SlI,12397
@@ -37,12 +37,12 @@ compressed_tensors/registry/__init__.py,sha256=FwLSNYqfIrb5JD_6OK_MT4_svvKTN_nEh
 compressed_tensors/registry/registry.py,sha256=fxjOjh2wklCvJhQxwofdy-zV8q7MkQ85SLG77nml2iA,11890
 compressed_tensors/utils/__init__.py,sha256=rvbIJlvdKYn4iX7r3KP6peCbU5uyMzgxwhsQstLoMxQ,785
 compressed_tensors/utils/helpers.py,sha256=d3yP9ViQ8R3GzMHfohxNlaokzyrRuj2PyjxWAJZmSws,3156
-compressed_tensors/utils/offload.py,sha256=BL7_cNAHTKbSta179R5R4ASk6oXuZhTJDY4D_8Lv2OE,3717
+compressed_tensors/utils/offload.py,sha256=qAMwoFT3WEQ9nB_SegE12ob8ghDugddQseE6z4vpZEE,3900
 compressed_tensors/utils/permutations_24.py,sha256=kx6fsfDHebx94zsSzhXGyCyuC9sVyah6BUUir_StT28,2530
 compressed_tensors/utils/safetensors_load.py,sha256=0MheXwx1jeY12PeISppiSIZHs6rmN2YddwPpFb9V67I,8527
 compressed_tensors/utils/semi_structured_conversions.py,sha256=g1EZHzdv-ko7ufPX430dp7wE33o6FWJXuSP4zZydCu0,13488
-compressed_tensors_nightly-0.5.0.20240804.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors_nightly-0.5.0.20240804.dist-info/METADATA,sha256=H78U9c-Vx4toSi_Hxc8V4zEfgvofEROPHcA_ICVjeFs,5680
-compressed_tensors_nightly-0.5.0.20240804.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
-compressed_tensors_nightly-0.5.0.20240804.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors_nightly-0.5.0.20240804.dist-info/RECORD,,
+compressed_tensors_nightly-0.5.0.20240806.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors_nightly-0.5.0.20240806.dist-info/METADATA,sha256=h7Y9E6J9JViwHSrAt7Ir3oBGRJ3CbXbQHiraPnonmNE,5680
+compressed_tensors_nightly-0.5.0.20240806.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
+compressed_tensors_nightly-0.5.0.20240806.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors_nightly-0.5.0.20240806.dist-info/RECORD,,