compressed-tensors 0.10.3a20250814__tar.gz → 0.11.0__tar.gz

{compressed_tensors-0.10.3a20250814/src/compressed_tensors.egg-info → compressed_tensors-0.11.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250814
+Version: 0.11.0
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250814 → compressed_tensors-0.11.0}/src/compressed_tensors/quantization/quant_scheme.py

@@ -57,6 +57,7 @@ class QuantizationScheme(BaseModel):
         inputs = model.input_activations
         outputs = model.output_activations
         weights = model.weights
+        format = model.format
 
         if inputs is not None:
             if inputs.actorder is not None:
@@ -66,6 +67,11 @@ class QuantizationScheme(BaseModel):
             if outputs.actorder is not None:
                 raise ValueError("Cannot apply actorder to output activations")
 
+        if format == CompressionFormat.mixed_precision.value:
+            raise ValueError(
+                "mixed-precision cannot be set as a format for a QuantizationScheme"
+            )
+
         if (
             inputs
             and weights

{compressed_tensors-0.10.3a20250814 → compressed_tensors-0.11.0}/src/compressed_tensors/utils/match.py

@@ -15,7 +15,7 @@
 import logging
 import re
 from collections.abc import Generator
-from typing import Iterable, Mapping, Optional, Tuple
+from typing import Iterable, List, Mapping, Optional, Tuple, Union
 
 import torch
 from compressed_tensors.utils.internal import InternalModule
@@ -57,10 +57,10 @@ def match_named_modules(
     unmatched_targets = set(targets)
     for name, module in model.named_modules():
         for target in targets:
-            if is_match(name, module, target, fused):
+            if is_match(name, module, target, fused=fused):
                 unmatched_targets -= {target}
 
-                if not any(is_match(name, module, ign, fused) for ign in ignore):
+                if not is_match(name, module, ignore, fused=fused):
                     yield name, module
 
     if warn_on_fail:
@@ -155,9 +155,7 @@ def match_modules_set(
     for name, module in model.named_modules():
         # match until we get a full set
         for target in targets:
-            if is_match(name, module, target) and not any(
-                is_match(name, module, ign) for ign in ignore
-            ):
+            if is_match(name, module, target, ignore):
                 if matches[target] is not None:
                     raise ValueError(f"Matched a {target} twice before completing set")
                 matches[target] = module
@@ -176,7 +174,8 @@ def match_modules_set(
 def is_match(
     name: str,
     module: torch.nn.Module,
-    target: str,
+    targets: Union[str, Iterable[str]],
+    ignore: Union[str, Iterable[str]] = tuple(),
     fused: Optional[FusedMappping] = None,
 ) -> bool:
     """
@@ -198,8 +197,17 @@ def is_match(
     :fused: optional mapping from suffixes of fused modules to the suffixes of their
         corresponding shards
     """
+    targets = [targets] if isinstance(targets, str) else targets
+    ignore = [ignore] if isinstance(ignore, str) else ignore
+
     return not isinstance(module, InternalModule) and (
-        _match_name(name, target, fused) or _match_class(module, target)
+        any(
+            _match_name(name, target, fused) or _match_class(module, target)
+            for target in targets
+        )
+        and not any(
+            _match_name(name, ign, fused) or _match_class(module, ign) for ign in ignore
+        )
     )
 
 

{compressed_tensors-0.10.3a20250814 → compressed_tensors-0.11.0}/src/compressed_tensors/utils/offload.py

@@ -579,6 +579,7 @@ def remove_dispatch(module: torch.nn.Module) -> torch.nn.Module:
     remove_hook_from_module(module, recurse=True)
     if hasattr(module, "hf_device_map"):
         delattr(module, "hf_device_map")
+    module.to("cpu")
 
     return module
 

{compressed_tensors-0.10.3a20250814 → compressed_tensors-0.11.0}/src/compressed_tensors/version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.10.3.a20250814'
-__version_tuple__ = version_tuple = (0, 10, 3)
+__version__ = version = '0.11.0'
+__version_tuple__ = version_tuple = (0, 11, 0)

{compressed_tensors-0.10.3a20250814 → compressed_tensors-0.11.0/src/compressed_tensors.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compressed-tensors
-Version: 0.10.3a20250814
+Version: 0.11.0
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed_tensors-0.10.3a20250814 → compressed_tensors-0.11.0}/tests/test_utils/test_match.py

@@ -201,14 +201,20 @@ class TestIsMatch:
             "gate_up_proj": ["gate_proj", "up_proj"],
         }
 
-        assert is_match("dummy.qkv_proj", linear, "re:.*q_proj", mapping) == True
-        assert is_match("dummy.qkv_proj", linear, "re:.*k_proj", mapping) == True
-        assert is_match("dummy.qkv_proj", linear, "re:.*v_proj", mapping) == True
-        assert is_match("dummy.qkv_proj", linear, "Linear", mapping) == True
-
-        assert is_match("dummy.gate_up_proj", linear, "re:.*gate_proj", mapping) == True
-        assert is_match("dummy.gate_up_proj", linear, "re:.*up_proj", mapping) == True
-        assert is_match("dummy.gate_up_proj", linear, "Linear", mapping) == True
+        assert is_match("dummy.qkv_proj", linear, "re:.*q_proj", fused=mapping) == True
+        assert is_match("dummy.qkv_proj", linear, "re:.*k_proj", fused=mapping) == True
+        assert is_match("dummy.qkv_proj", linear, "re:.*v_proj", fused=mapping) == True
+        assert is_match("dummy.qkv_proj", linear, "Linear", fused=mapping) == True
+
+        assert (
+            is_match("dummy.gate_up_proj", linear, "re:.*gate_proj", fused=mapping)
+            == True
+        )
+        assert (
+            is_match("dummy.gate_up_proj", linear, "re:.*up_proj", fused=mapping)
+            == True
+        )
+        assert is_match("dummy.gate_up_proj", linear, "Linear", fused=mapping) == True
 
 
 class TestMatchNamedModules: