compressed-tensors-nightly 0.7.1.20241101__py3-none-any.whl → 0.7.1.20241102__py3-none-any.whl

compressed_tensors/quantization/lifecycle/forward.py

@@ -14,7 +14,7 @@
 
 from functools import wraps
 from math import ceil
-from typing import Callable, Optional
+from typing import Optional
 
 import torch
 from compressed_tensors.quantization.quant_args import (
@@ -28,7 +28,7 @@ from compressed_tensors.quantization.utils import (
     calculate_range,
     compute_dynamic_scales_and_zp,
 )
-from compressed_tensors.utils import safe_permute, update_parameter_data
+from compressed_tensors.utils import safe_permute
 from torch.nn import Module
 
 

compressed_tensors/quantization/quant_args.py

@@ -114,12 +114,6 @@ class QuantizationArgs(BaseModel, use_enum_values=True):
         """
         :return: torch quantization FakeQuantize built based on these QuantizationArgs
         """
-
-        # No observer required for the dynamic case
-        if self.dynamic:
-            self.observer = None
-            return self.observer
-
         return self.observer
 
     @field_validator("type", mode="before")
@@ -203,6 +197,7 @@ class QuantizationArgs(BaseModel, use_enum_values=True):
                 "activation ordering"
             )
 
+        # infer observer w.r.t. dynamic
         if dynamic:
             if strategy not in (
                 QuantizationStrategy.TOKEN,
@@ -214,18 +209,19 @@ class QuantizationArgs(BaseModel, use_enum_values=True):
                     "quantization",
                 )
             if observer is not None:
-                warnings.warn(
-                    "No observer is used for dynamic quantization, setting to None"
-                )
-                model.observer = None
+                if observer != "memoryless":  # avoid annoying users with old configs
+                    warnings.warn(
+                        "No observer is used for dynamic quantization, setting to None"
+                    )
+                observer = None
 
-        # if we have not set an observer and we
-        # are running static quantization, use minmax
-        if not observer and not dynamic:
-            model.observer = "minmax"
+        elif observer is None:
+            # default to minmax for non-dynamic cases
+            observer = "minmax"
 
         # write back modified values
         model.strategy = strategy
+        model.observer = observer
         return model
 
     def pytorch_dtype(self) -> torch.dtype:

compressed_tensors/quantization/utils/helpers.py

@@ -233,7 +233,6 @@ def iter_named_leaf_modules(model: Module) -> Generator[Tuple[str, Module], None
                 named_children, children = zip(*list(submodule.named_children()))
             has_non_observer_children = False
             for i in range(len(children)):
-                child = children[i]
                 child_name = named_children[i]
 
                 if "observer" not in child_name:
@@ -268,7 +267,6 @@ def iter_named_quantizable_modules(
                 has_non_observer_children = False
                 for i in range(len(children)):
                     child_name = named_children[i]
-                    child = children[i]
 
                     if "observer" not in child_name:
                         has_non_observer_children = True

compressed_tensors/registry/registry.py

@@ -258,7 +258,7 @@ def get_from_registry(
         retrieved_value = _import_and_get_value_from_module(module_path, value_name)
     else:
         # look up name in alias registry
-        name = _ALIAS_REGISTRY[parent_class].get(name)
+        name = _ALIAS_REGISTRY[parent_class].get(name, name)
         # look up name in registry
         retrieved_value = _REGISTRY[parent_class].get(name)
         if retrieved_value is None:

{compressed_tensors_nightly-0.7.1.20241101.dist-info → compressed_tensors_nightly-0.7.1.20241102.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: compressed-tensors-nightly
-Version: 0.7.1.20241101
+Version: 0.7.1.20241102
 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-0.7.1.20241101.dist-info → compressed_tensors_nightly-0.7.1.20241102.dist-info}/RECORD

@@ -23,19 +23,19 @@ compressed_tensors/config/sparse_bitmask.py,sha256=pZUboRNZTu6NajGOQEFExoPknak5y
 compressed_tensors/linear/__init__.py,sha256=fH6rjBYAxuwrTzBTlTjTgCYNyh6TCvCqajCz4Im4YrA,617
 compressed_tensors/linear/compressed_linear.py,sha256=0jTTf6XxOAjAYs3tvFtgiNMAO4W10sSeR-pdH2M413g,3218
 compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
-compressed_tensors/quantization/quant_args.py,sha256=VNt6hKGGdjFsy0A6e5biQmW0EhkrTZvdEVuKhfd0k1M,8754
+compressed_tensors/quantization/quant_args.py,sha256=osjNwCSB6tcyH9Qeg5sHEiB-bHyi3XJ8TzkGVJuGTc4,8711
 compressed_tensors/quantization/quant_config.py,sha256=NCiMvUMnnz5kTyAkDylxjtEGQnjgsIYIeNR2zyHEdTQ,10371
 compressed_tensors/quantization/quant_scheme.py,sha256=5ggPz5sqEfTUgvJJeiPIINA74QtO-08hb3szsm7UHGE,6000
 compressed_tensors/quantization/lifecycle/__init__.py,sha256=_uItzFWusyV74Zco_pHLOTdE9a83cL-R-ZdyQrBkIyw,772
 compressed_tensors/quantization/lifecycle/apply.py,sha256=pdCqxXnVw7HoDDanaOtek13g8x_nb54CBUlfuMdhFG4,14993
 compressed_tensors/quantization/lifecycle/compressed.py,sha256=Fj9n66IN0EWsOAkBHg3O0GlOQpxstqjCcs0ttzMXrJ0,2296
-compressed_tensors/quantization/lifecycle/forward.py,sha256=WSHSBsOJqtP5WxYYd1sU3WfwTZ8zbcs-dOf4hUvmKME,12974
+compressed_tensors/quantization/lifecycle/forward.py,sha256=QPL6-vKOFuKdKIEsVqMhsw4x552Jpm2sqO0oeChbnrM,12941
 compressed_tensors/quantization/lifecycle/helpers.py,sha256=C0mhy2vJ0fCjVeN4kFNhw8Eq1wkteBGHiZ36RVLThRY,944
 compressed_tensors/quantization/lifecycle/initialize.py,sha256=C41hKA5VANyEwkB5FxzEn3Z0Da5tfxF1I07P8rUcyS0,8537
 compressed_tensors/quantization/utils/__init__.py,sha256=VdtEmP0bvuND_IGQnyqUPc5lnFp-1_yD7StKSX4x80w,656
-compressed_tensors/quantization/utils/helpers.py,sha256=FfFSHs67azZU30ekSax9jEeX9flps2WN8UMC8tWbwbc,14176
+compressed_tensors/quantization/utils/helpers.py,sha256=DBP-sGRpGAY01K0LFE7qqonNj4hkTYL_mXrMs2LtAD8,14100
 compressed_tensors/registry/__init__.py,sha256=FwLSNYqfIrb5JD_6OK_MT4_svvKTN_nEhpgQlQvGbjI,658
-compressed_tensors/registry/registry.py,sha256=fxjOjh2wklCvJhQxwofdy-zV8q7MkQ85SLG77nml2iA,11890
+compressed_tensors/registry/registry.py,sha256=vRcjVB1ITfSbfYUaGndBBmqhip_5vsS62weorVg0iXo,11896
 compressed_tensors/utils/__init__.py,sha256=gS4gSU2pwcAbsKj-6YMaqhm25udFy6ISYaWBf-myRSM,808
 compressed_tensors/utils/helpers.py,sha256=hWGIR0W7ENHwdC7wW2SQJJiCF9-xOu_u3fY2RzLyYg4,4101
 compressed_tensors/utils/offload.py,sha256=d9q8LNe8HyF8tOjgjA7QGLD3HRysmNp0d8eBbdqBgIM,4089
@@ -43,8 +43,8 @@ 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=m08ANVuTBxQdoa6LufDgcNJ7wCLDJolyZljB8VEybAU,8578
 compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
-compressed_tensors_nightly-0.7.1.20241101.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors_nightly-0.7.1.20241101.dist-info/METADATA,sha256=ZGzpjiwBhKjCRZskb8-8aS_R6iFAMSBsmFztEKGbbIE,6799
-compressed_tensors_nightly-0.7.1.20241101.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
-compressed_tensors_nightly-0.7.1.20241101.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors_nightly-0.7.1.20241101.dist-info/RECORD,,
+compressed_tensors_nightly-0.7.1.20241102.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors_nightly-0.7.1.20241102.dist-info/METADATA,sha256=pQ8FXKctjUHKkisrXYyeDUuunknVPkjHnHvS-uJ89oI,6799
+compressed_tensors_nightly-0.7.1.20241102.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
+compressed_tensors_nightly-0.7.1.20241102.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors_nightly-0.7.1.20241102.dist-info/RECORD,,