compressed-tensors 0.6.0__py3-none-any.whl → 0.7.0__py3-none-any.whl

compressed_tensors/quantization/quant_config.py

@@ -24,7 +24,7 @@ from compressed_tensors.quantization.quant_scheme import (
 from compressed_tensors.quantization.utils import (
     calculate_compression_ratio,
     is_module_quantized,
-    iter_named_leaf_modules,
+    iter_named_quantizable_modules,
     module_type,
     parse_out_kv_cache_args,
 )
@@ -177,7 +177,9 @@ class QuantizationConfig(BaseModel):
         quantization_status = None
         ignore = {}
         quantization_type_names = set()
-        for name, submodule in iter_named_leaf_modules(model):
+        for name, submodule in iter_named_quantizable_modules(
+            model, include_children=True, include_attn=True
+        ):
             layer_type = module_type(submodule)
             if not is_module_quantized(submodule):
                 if layer_type not in ignore:
@@ -199,6 +201,13 @@ class QuantizationConfig(BaseModel):
         if len(quant_scheme_to_layers) == 0:  # No quantized layers
             return None
 
+        # kv-cache only, no weight/activation quantization
+        if (
+            len(quantization_type_names) == 1
+            and "attention" in list(quantization_type_names)[0].lower()
+        ):
+            quantization_type_names.add("Linear")
+
         # clean up ignore list, we can leave out layers types if none of the
         # instances are quantized
         consolidated_ignore = []
@@ -241,6 +250,9 @@ class QuantizationConfig(BaseModel):
         )
 
     def requires_calibration_data(self):
+        if self.kv_cache_scheme is not None:
+            return True
+
         for _, scheme in self.config_groups.items():
             if scheme.input_activations is not None:
                 if not scheme.input_activations.dynamic:

compressed_tensors/quantization/quant_scheme.py

@@ -108,7 +108,7 @@ def is_preset_scheme(name: str) -> bool:
 UNQUANTIZED = dict()
 
 # 8 bit integer weights and 8 bit activations quantization
-W8A8 = dict(
+INT8_W8A8 = dict(
     weights=QuantizationArgs(
         num_bits=8,
         type=QuantizationType.INT,
@@ -149,7 +149,7 @@ W4A16 = dict(
 )
 
 # 4 bit integer weights and 8 bit activations quantization
-W4A8 = dict(
+INT8_W4A8 = dict(
     weights=QuantizationArgs(
         num_bits=4,
         type=QuantizationType.INT,
@@ -210,8 +210,9 @@ PRESET_SCHEMES = {
     "W8A16": W8A16,
     "W4A16": W4A16,
     # Integer weight and activation schemes
-    "W8A8": W8A8,
-    "W4A8": W4A8,
+    "W8A8": INT8_W8A8,
+    "INT8": INT8_W8A8,  # alias for W8A8
+    "W4A8": INT8_W4A8,
     # Float weight and activation schemes
     "FP8": FP8,
     "FP8_DYNAMIC": FP8_DYNAMIC,

compressed_tensors/quantization/utils/helpers.py

@@ -13,8 +13,7 @@
 # limitations under the License.
 
 import logging
-import re
-from typing import List, Optional, Tuple
+from typing import Generator, List, Optional, Tuple
 
 import torch
 from compressed_tensors.quantization.observers.base import Observer
@@ -28,7 +27,6 @@ __all__ = [
     "infer_quantization_status",
     "is_module_quantized",
     "is_model_quantized",
-    "iter_named_leaf_modules",
     "module_type",
     "calculate_compression_ratio",
     "get_torch_bit_depth",
@@ -36,9 +34,14 @@ __all__ = [
     "parse_out_kv_cache_args",
     "KV_CACHE_TARGETS",
     "is_kv_cache_quant_scheme",
+    "iter_named_leaf_modules",
+    "iter_named_quantizable_modules",
 ]
 
-KV_CACHE_TARGETS = ["re:.*k_proj", "re:.*v_proj"]
+# target the self_attn layer
+# QuantizedKVParameterCache is responsible for obtaining the k_scale and v_scale
+KV_CACHE_TARGETS = ["re:.*self_attn$"]
+
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 
@@ -106,11 +109,10 @@ def module_type(module: Module) -> str:
     return type(module).__name__
 
 
-def iter_named_leaf_modules(model: Module) -> Tuple[str, Module]:
+def iter_named_leaf_modules(model: Module) -> Generator[Tuple[str, Module], None, None]:
     """
     Yields modules that do not have any submodules except observers. The observers
     themselves are not yielded
-
     :param model: model to get leaf modules of
     :returns: generator tuple of (name, leaf_submodule)
     """
@@ -128,6 +130,37 @@ def iter_named_leaf_modules(model: Module) -> Tuple[str, Module]:
                 yield name, submodule
 
 
+def iter_named_quantizable_modules(
+    model: Module, include_children: bool = True, include_attn: bool = False
+) -> Generator[Tuple[str, Module], None, None]:
+    """
+    Yield name and submodule of
+    - leaf modules, set by include_children
+    - attention modyles, set by include_attn
+
+    :param model: model to get leaf modules of
+    :param include_children: flag to get the leaf modules
+    :param inlcude_attn: flag to get the attention modules
+    :returns: generator tuple of (name, submodule)
+    """
+    for name, submodule in model.named_modules():
+        if include_children:
+            children = list(submodule.children())
+            if len(children) == 0 and not isinstance(submodule, Observer):
+                yield name, submodule
+            else:
+                has_non_observer_children = False
+                for child in children:
+                    if not isinstance(child, Observer):
+                        has_non_observer_children = True
+
+                if not has_non_observer_children:
+                    yield name, submodule
+        if include_attn:
+            if name.endswith("self_attn"):
+                yield name, submodule
+
+
 def get_torch_bit_depth(value: torch.Tensor) -> int:
     """
     Determine the number of bits used to represent the dtype of a tensor
@@ -204,19 +237,11 @@ def is_kv_cache_quant_scheme(scheme: QuantizationScheme) -> bool:
     :param scheme: The QuantizationScheme to investigate
     :return: boolean flag
     """
-    if len(scheme.targets) == 1:
-        # match on the KV_CACHE_TARGETS regex pattern
-        # if there is only one target
-        is_match_targets = any(
-            [re.match(pattern[3:], scheme.targets[0]) for pattern in KV_CACHE_TARGETS]
-        )
-    else:
-        # match on the exact KV_CACHE_TARGETS
-        # if there are multiple targets
-        is_match_targets = set(KV_CACHE_TARGETS) == set(scheme.targets)
+    for target in scheme.targets:
+        if target in KV_CACHE_TARGETS:
+            return True
 
-    is_match_output_activations = scheme.output_activations is not None
-    return is_match_targets and is_match_output_activations
+    return False
 
 
 def parse_out_kv_cache_args(

compressed_tensors/utils/helpers.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Optional
+from typing import Any, Optional
 
 import torch
 from transformers import AutoConfig
@@ -23,6 +23,7 @@ __all__ = [
     "fix_fsdp_module_name",
     "tensor_follows_mask_structure",
     "replace_module",
+    "is_compressed_tensors_config",
 ]
 
 FSDP_WRAPPER_NAME = "_fsdp_wrapped_module"
@@ -103,3 +104,18 @@ def replace_module(model: torch.nn.Module, name: str, new_module: torch.nn.Modul
         parent = model
         child_name = name
     setattr(parent, child_name, new_module)
+
+
+def is_compressed_tensors_config(compression_config: Any) -> bool:
+    """
+    Returns True if CompressedTensorsConfig is available from transformers and
+    compression_config is an instance of CompressedTensorsConfig
+
+    See: https://github.com/huggingface/transformers/pull/31704
+    """
+    try:
+        from transformers.utils.quantization_config import CompressedTensorsConfig
+
+        return isinstance(compression_config, CompressedTensorsConfig)
+    except ImportError:
+        return False

compressed_tensors/version.py

@@ -17,7 +17,7 @@ Functionality for storing and setting the version info for SparseML
 """
 
 
-version_base = "0.6.0"
+version_base = "0.7.0"
 is_release = True  # change to True to set the generated version as a release version
 
 

{compressed_tensors-0.6.0.dist-info → compressed_tensors-0.7.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: compressed-tensors
-Version: 0.6.0
+Version: 0.7.0
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

compressed_tensors-0.7.0.dist-info/RECORD

@@ -0,0 +1,59 @@
+compressed_tensors/__init__.py,sha256=UtKmifNeBCSE2TZSAfduVNNzHY-3V7bLjZ7n7RuXLOE,812
+compressed_tensors/base.py,sha256=73HYH7HY7O2roC89yG_piPFnZwrBfn_i7HmKl90SKc0,875
+compressed_tensors/version.py,sha256=RTYptXdV8f4QbYCRQ13eGeEsq4grNJs6EXgejoZl9EE,1585
+compressed_tensors/compressors/__init__.py,sha256=smSygTSfcfuujRrAXDc6uZm4L_ccV1tWZewqVnOb4lM,825
+compressed_tensors/compressors/base.py,sha256=D9TNwQcjanDiAHODPbg8JUqc66e3j50rctY7A708NEs,6743
+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=XJgPsq8KiDfiR4e8bSI38lmoOd2ApqRk1aPcXS2obqY,15600
+compressed_tensors/compressors/quantized_compressors/__init__.py,sha256=09UJq68Pht6Bf-4iP9xYl3tetKsncNPHD8IAGbePsr4,714
+compressed_tensors/compressors/quantized_compressors/base.py,sha256=K1KOnS6Y8nUA1-HN7VhyfsDc01nilW0WfXMUhuD-l8w,5954
+compressed_tensors/compressors/quantized_compressors/naive_quantized.py,sha256=Mmfr-hap-4zw7CzE1mXi0UirknqGidNxw38GGWVgTqM,4916
+compressed_tensors/compressors/quantized_compressors/pack_quantized.py,sha256=9H8UrG5v1GRtslLjOEiUM2dnyxJnR-HJmlsFezQs_r0,7706
+compressed_tensors/compressors/sparse_compressors/__init__.py,sha256=i2TESH27l7KXeOhJ6hShIoI904XX96l-cRQiMR6MAaU,704
+compressed_tensors/compressors/sparse_compressors/base.py,sha256=Ua4rUSGyucEs-YJI5z3oIUF-zqQLrFsQ9f-qKasEdUM,4410
+compressed_tensors/compressors/sparse_compressors/dense.py,sha256=lSKNWRx6H7aUqaJj1j4qbXk8Gkm1UohbnvW1Rvq6Ra4,1284
+compressed_tensors/compressors/sparse_compressors/sparse_bitmask.py,sha256=4fKwCG7ZM8mUtSnjPvubzEHl-mTnxMzwjmcs7L43WLY,6622
+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=akqE7eW8CLTslpWRxERaZ8R0TSm1lS7D1bgZXKL0xi8,9427
+compressed_tensors/config/__init__.py,sha256=ZBqWn3r6ku1qfmlHHYp0mQueY0i7Pwhr9rbQk9dDlMc,704
+compressed_tensors/config/base.py,sha256=BNTFKy12isY7qblwxdi_R1f00EzgrNOXLrfxqLCPT8w,1903
+compressed_tensors/config/dense.py,sha256=NgSxnFCnckU9-iunxEaqiFwqgdO7YYxlWKR74jNbjks,1317
+compressed_tensors/config/sparse_bitmask.py,sha256=pZUboRNZTu6NajGOQEFExoPknak5ynVAUeiiYpS1Gt8,1308
+compressed_tensors/linear/__init__.py,sha256=fH6rjBYAxuwrTzBTlTjTgCYNyh6TCvCqajCz4Im4YrA,617
+compressed_tensors/linear/compressed_linear.py,sha256=0jTTf6XxOAjAYs3tvFtgiNMAO4W10sSeR-pdH2M413g,3218
+compressed_tensors/quantization/__init__.py,sha256=nWP_fsl6Nn0ksEgZPzerGiETdvF-ZfNwPnwGlRiR5pY,805
+compressed_tensors/quantization/cache.py,sha256=vnBB5zasO_XpHomZvzUPVVbzyCz2VgebsHePm0kANzY,6831
+compressed_tensors/quantization/quant_args.py,sha256=73KevZXHyrkMCT_3CxbYHz70fI3i-wcF8NvN0wsBPK4,8271
+compressed_tensors/quantization/quant_config.py,sha256=NCiMvUMnnz5kTyAkDylxjtEGQnjgsIYIeNR2zyHEdTQ,10371
+compressed_tensors/quantization/quant_scheme.py,sha256=uFgp6ECU6ZkHWkeKlAVAzZTLDbrTrzPSPrY23eJluaw,5931
+compressed_tensors/quantization/lifecycle/__init__.py,sha256=MXE2E7GfIfRRfhrdGy2Og3AZOz5N59B0ZGFcsD89y6c,821
+compressed_tensors/quantization/lifecycle/apply.py,sha256=czaayvpeUYyWRJhO_klffw6esptOgA9sBKL5TWQcRdw,15805
+compressed_tensors/quantization/lifecycle/calibration.py,sha256=IuLeRkVQPrMxkMcIjr4OMFlIUMHkqjH4qAxC2KiUBGw,2673
+compressed_tensors/quantization/lifecycle/compressed.py,sha256=Fj9n66IN0EWsOAkBHg3O0GlOQpxstqjCcs0ttzMXrJ0,2296
+compressed_tensors/quantization/lifecycle/forward.py,sha256=eLup6QDRUUp_Ozcas7RDRLIXBWjFbxn5gWbcAIJEGlw,15715
+compressed_tensors/quantization/lifecycle/frozen.py,sha256=NiJw7NP7pcT6idWFa8vksgiLoT8oQ975e57S4QfD2QQ,1874
+compressed_tensors/quantization/lifecycle/helpers.py,sha256=C0mhy2vJ0fCjVeN4kFNhw8Eq1wkteBGHiZ36RVLThRY,944
+compressed_tensors/quantization/lifecycle/initialize.py,sha256=4_YG7jKl7d2-Cy58pOkMtInFRhvYahxYchesWMPxPVM,8862
+compressed_tensors/quantization/observers/__init__.py,sha256=4Sa7rqi5RB_S5bPO8KmncETiqDsoMBhwP37arlQym8s,764
+compressed_tensors/quantization/observers/base.py,sha256=5ovQicWPYHjIxr6-EkQ4lgOX0PpI9g23iSzKpxjM1Zg,8420
+compressed_tensors/quantization/observers/helpers.py,sha256=s_A23Qa_BLfOdHJCN5bm-qPWkhjjj_RIVrhSp1Y9Dtk,4211
+compressed_tensors/quantization/observers/memoryless.py,sha256=jH_c6K3gxf4W3VNXQ7tbnP-J_86QTrEfjBn6Kh1C-H8,2165
+compressed_tensors/quantization/observers/min_max.py,sha256=sQXqU3z-voxIDfR_9mQzwQUflZj2sASm_G8CYaXntFw,3865
+compressed_tensors/quantization/observers/mse.py,sha256=Aeh-253Vbab1F8cYuBiGNn4OXWJ67wXQ_JVfl3mu2a8,6034
+compressed_tensors/quantization/utils/__init__.py,sha256=VdtEmP0bvuND_IGQnyqUPc5lnFp-1_yD7StKSX4x80w,656
+compressed_tensors/quantization/utils/helpers.py,sha256=y4LEyC2oUd876ZMdALWKGH3Ct5EgBJZV4id_NUjTGH8,9531
+compressed_tensors/registry/__init__.py,sha256=FwLSNYqfIrb5JD_6OK_MT4_svvKTN_nEhpgQlQvGbjI,658
+compressed_tensors/registry/registry.py,sha256=fxjOjh2wklCvJhQxwofdy-zV8q7MkQ85SLG77nml2iA,11890
+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
+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=m08ANVuTBxQdoa6LufDgcNJ7wCLDJolyZljB8VEybAU,8578
+compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
+compressed_tensors-0.7.0.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+compressed_tensors-0.7.0.dist-info/METADATA,sha256=Lgcl4rU8ifo0PY-FrurFApAkuTD9HBeJohuULjVqebs,6782
+compressed_tensors-0.7.0.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
+compressed_tensors-0.7.0.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
+compressed_tensors-0.7.0.dist-info/RECORD,,

compressed_tensors-0.6.0.dist-info/RECORD

@@ -1,52 +0,0 @@
-compressed_tensors/__init__.py,sha256=SV1csvHUVCd8kHXz6UDZim1HZ_fAVG3vfk-j_4Bb6hY,789
-compressed_tensors/base.py,sha256=Mq4mfVQcJhNpha-BXzpOfpmFIdl01o09BJE7D2oQ_00,796
-compressed_tensors/version.py,sha256=qWZbaQsbr0_5Y1qtWGstNPljw-ggMLBGZBvSKfzB9pw,1585
-compressed_tensors/compressors/__init__.py,sha256=wmX4VnkUTS63xBwK5-6w8FP78bNZpcdcqvf2KOEC5E4,1133
-compressed_tensors/compressors/base.py,sha256=NfVkhq6PRiq2cvAXaUXLoqC_nVYWdSrkE12c9AXYSMo,9956
-compressed_tensors/compressors/dense.py,sha256=xcWECjcRY4INN6jC7vHx5wvUX3NmnKlxA9SVE1A6m2Q,1267
-compressed_tensors/compressors/helpers.py,sha256=k9avlkmeYj6vkOAvl-MgcixtP7ib24SCfhzZ-RusXfw,5403
-compressed_tensors/compressors/marlin_24.py,sha256=e7fGUyZbjUpA5VUMCPxqcYPGNiwoDKupHJaXWCoVKRw,9410
-compressed_tensors/compressors/model_compressor.py,sha256=gI6KKtH3eeWi2540Ayx-4bg9o8qjrvxlF4Gd_sqltGA,16678
-compressed_tensors/compressors/naive_quantized.py,sha256=z3h3ca5xKCN69mahutxcbzdv-OysiaxaM8P-Qum6zUQ,4823
-compressed_tensors/compressors/pack_quantized.py,sha256=27RVmJ2wg2dvCoawj407HSmKT3VPGJ6ujAMHlT26WlI,7571
-compressed_tensors/compressors/sparse_bitmask.py,sha256=kiDwBlFV0sJGLcIdDYxIiuF64ccgwDfqq1hWRQThYDc,8647
-compressed_tensors/config/__init__.py,sha256=ZBqWn3r6ku1qfmlHHYp0mQueY0i7Pwhr9rbQk9dDlMc,704
-compressed_tensors/config/base.py,sha256=BNTFKy12isY7qblwxdi_R1f00EzgrNOXLrfxqLCPT8w,1903
-compressed_tensors/config/dense.py,sha256=NgSxnFCnckU9-iunxEaqiFwqgdO7YYxlWKR74jNbjks,1317
-compressed_tensors/config/sparse_bitmask.py,sha256=pZUboRNZTu6NajGOQEFExoPknak5ynVAUeiiYpS1Gt8,1308
-compressed_tensors/linear/__init__.py,sha256=fH6rjBYAxuwrTzBTlTjTgCYNyh6TCvCqajCz4Im4YrA,617
-compressed_tensors/linear/compressed_linear.py,sha256=G0gEFfxLAUsgRcnfSV-PKz1ZBNTVokOauOoup7SE1mw,3210
-compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
-compressed_tensors/quantization/quant_args.py,sha256=CmyVtjJeHlqCW-7R5Z7tIw6lXUrzCX6Y9bwgmMxEudY,8069
-compressed_tensors/quantization/quant_config.py,sha256=NpVu8YJ4Xw2pIQW_PGaNaml8kx1bUnxkvb0jBYWbKdE,9971
-compressed_tensors/quantization/quant_scheme.py,sha256=HmR1DcFZcjuqX7KHUYI0NFXsCIzJ8sxFGH6zhYuHmEs,5870
-compressed_tensors/quantization/lifecycle/__init__.py,sha256=MXE2E7GfIfRRfhrdGy2Og3AZOz5N59B0ZGFcsD89y6c,821
-compressed_tensors/quantization/lifecycle/apply.py,sha256=uftWFunr_CpCZM_qWfo2O1USXKB2qSYD1pBJsO8BuCU,15285
-compressed_tensors/quantization/lifecycle/calibration.py,sha256=PlS_EqCOPqJD3QKuLPXO9AOtDzXtQWvEBTynFv-FFVw,2698
-compressed_tensors/quantization/lifecycle/compressed.py,sha256=Fj9n66IN0EWsOAkBHg3O0GlOQpxstqjCcs0ttzMXrJ0,2296
-compressed_tensors/quantization/lifecycle/forward.py,sha256=PljD9pzATILEOiC3ZdHUTsfSbZdAa6iSIxWmvAHLG9I,13688
-compressed_tensors/quantization/lifecycle/frozen.py,sha256=h1XYt89MouBTf3jTYLG_6OdFxIu5q2N8tPjsy6J4E6Y,1726
-compressed_tensors/quantization/lifecycle/helpers.py,sha256=TmLY_G5VP_Fg2Ywio_dxoHRTxOKZdT7_aG5S9WtD4zI,2424
-compressed_tensors/quantization/lifecycle/initialize.py,sha256=S5Kwy16Da8WUIIpa1xVKc72MijJ5C_rqM6JjanZ7MGk,7133
-compressed_tensors/quantization/observers/__init__.py,sha256=4Sa7rqi5RB_S5bPO8KmncETiqDsoMBhwP37arlQym8s,764
-compressed_tensors/quantization/observers/base.py,sha256=5ovQicWPYHjIxr6-EkQ4lgOX0PpI9g23iSzKpxjM1Zg,8420
-compressed_tensors/quantization/observers/helpers.py,sha256=s_A23Qa_BLfOdHJCN5bm-qPWkhjjj_RIVrhSp1Y9Dtk,4211
-compressed_tensors/quantization/observers/memoryless.py,sha256=jH_c6K3gxf4W3VNXQ7tbnP-J_86QTrEfjBn6Kh1C-H8,2165
-compressed_tensors/quantization/observers/min_max.py,sha256=sQXqU3z-voxIDfR_9mQzwQUflZj2sASm_G8CYaXntFw,3865
-compressed_tensors/quantization/observers/mse.py,sha256=Aeh-253Vbab1F8cYuBiGNn4OXWJ67wXQ_JVfl3mu2a8,6034
-compressed_tensors/quantization/utils/__init__.py,sha256=VdtEmP0bvuND_IGQnyqUPc5lnFp-1_yD7StKSX4x80w,656
-compressed_tensors/quantization/utils/helpers.py,sha256=pwvU613XRvMDtI5b39II5jukBl5OUCqoX0ofVRpOFRY,8633
-compressed_tensors/registry/__init__.py,sha256=FwLSNYqfIrb5JD_6OK_MT4_svvKTN_nEhpgQlQvGbjI,658
-compressed_tensors/registry/registry.py,sha256=fxjOjh2wklCvJhQxwofdy-zV8q7MkQ85SLG77nml2iA,11890
-compressed_tensors/utils/__init__.py,sha256=gS4gSU2pwcAbsKj-6YMaqhm25udFy6ISYaWBf-myRSM,808
-compressed_tensors/utils/helpers.py,sha256=bh4G8mj_YCRf8Bo2FQ9FkIIZXY8xqqPjckNnVYB0gBA,3557
-compressed_tensors/utils/offload.py,sha256=d9q8LNe8HyF8tOjgjA7QGLD3HRysmNp0d8eBbdqBgIM,4089
-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=m08ANVuTBxQdoa6LufDgcNJ7wCLDJolyZljB8VEybAU,8578
-compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
-compressed_tensors-0.6.0.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-compressed_tensors-0.6.0.dist-info/METADATA,sha256=opV5oXtNn7EmogPQaMiv9ddOLHv-XObt_NuUSwR5Bz4,6782
-compressed_tensors-0.6.0.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
-compressed_tensors-0.6.0.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
-compressed_tensors-0.6.0.dist-info/RECORD,,