compressed-tensors-nightly 0.6.0.20240926__py3-none-any.whl → 0.6.0.20240928__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- compressed_tensors/compressors/model_compressor.py +9 -4
- compressed_tensors/quantization/lifecycle/initialize.py +9 -2
- {compressed_tensors_nightly-0.6.0.20240926.dist-info → compressed_tensors_nightly-0.6.0.20240928.dist-info}/METADATA +1 -1
- {compressed_tensors_nightly-0.6.0.20240926.dist-info → compressed_tensors_nightly-0.6.0.20240928.dist-info}/RECORD +7 -7
- {compressed_tensors_nightly-0.6.0.20240926.dist-info → compressed_tensors_nightly-0.6.0.20240928.dist-info}/LICENSE +0 -0
- {compressed_tensors_nightly-0.6.0.20240926.dist-info → compressed_tensors_nightly-0.6.0.20240928.dist-info}/WHEEL +0 -0
- {compressed_tensors_nightly-0.6.0.20240926.dist-info → compressed_tensors_nightly-0.6.0.20240928.dist-info}/top_level.txt +0 -0
@@ -216,6 +216,11 @@ class ModelCompressor:
|
|
216
216
|
self.sparsity_compressor = None
|
217
217
|
self.quantization_compressor = None
|
218
218
|
|
219
|
+
|
220
|
+
if sparsity_config and sparsity_config.format == CompressionFormat.dense.value:
|
221
|
+
# ignore dense sparsity config
|
222
|
+
self.sparsity_config = None
|
223
|
+
|
219
224
|
if sparsity_config is not None:
|
220
225
|
self.sparsity_compressor = Compressor.load_from_registry(
|
221
226
|
sparsity_config.format, config=sparsity_config
|
@@ -306,16 +311,16 @@ class ModelCompressor:
|
|
306
311
|
with open(config_file_path, "r") as config_file:
|
307
312
|
config_data = json.load(config_file)
|
308
313
|
|
309
|
-
config_data[
|
314
|
+
config_data[QUANTIZATION_CONFIG_NAME] = {}
|
310
315
|
if self.quantization_config is not None:
|
311
316
|
quant_config_data = self.quantization_config.model_dump()
|
312
|
-
config_data[
|
317
|
+
config_data[QUANTIZATION_CONFIG_NAME] = quant_config_data
|
313
318
|
if self.sparsity_config is not None:
|
314
319
|
sparsity_config_data = self.sparsity_config.model_dump()
|
315
|
-
config_data[
|
320
|
+
config_data[QUANTIZATION_CONFIG_NAME][
|
316
321
|
SPARSITY_CONFIG_NAME
|
317
322
|
] = sparsity_config_data
|
318
|
-
config_data[
|
323
|
+
config_data[QUANTIZATION_CONFIG_NAME][
|
319
324
|
COMPRESSION_VERSION_NAME
|
320
325
|
] = compressed_tensors.__version__
|
321
326
|
|
@@ -75,7 +75,10 @@ def initialize_module_for_quantization(
|
|
75
75
|
|
76
76
|
if scheme.input_activations is not None:
|
77
77
|
_initialize_scale_zero_point_observer(
|
78
|
-
module,
|
78
|
+
module,
|
79
|
+
"input",
|
80
|
+
scheme.input_activations,
|
81
|
+
force_zero_point=force_zero_point,
|
79
82
|
)
|
80
83
|
if scheme.weights is not None:
|
81
84
|
if hasattr(module, "weight"):
|
@@ -83,7 +86,11 @@ def initialize_module_for_quantization(
|
|
83
86
|
if isinstance(module, torch.nn.Linear):
|
84
87
|
weight_shape = module.weight.shape
|
85
88
|
_initialize_scale_zero_point_observer(
|
86
|
-
module,
|
89
|
+
module,
|
90
|
+
"weight",
|
91
|
+
scheme.weights,
|
92
|
+
weight_shape=weight_shape,
|
93
|
+
force_zero_point=force_zero_point,
|
87
94
|
)
|
88
95
|
else:
|
89
96
|
_LOGGER.warning(
|
@@ -1,6 +1,6 @@
|
|
1
1
|
Metadata-Version: 2.1
|
2
2
|
Name: compressed-tensors-nightly
|
3
|
-
Version: 0.6.0.
|
3
|
+
Version: 0.6.0.20240928
|
4
4
|
Summary: Library for utilization of compressed safetensors of neural network models
|
5
5
|
Home-page: https://github.com/neuralmagic/compressed-tensors
|
6
6
|
Author: Neuralmagic, Inc.
|
@@ -6,7 +6,7 @@ compressed_tensors/compressors/base.py,sha256=NfVkhq6PRiq2cvAXaUXLoqC_nVYWdSrkE1
|
|
6
6
|
compressed_tensors/compressors/dense.py,sha256=xcWECjcRY4INN6jC7vHx5wvUX3NmnKlxA9SVE1A6m2Q,1267
|
7
7
|
compressed_tensors/compressors/helpers.py,sha256=k9avlkmeYj6vkOAvl-MgcixtP7ib24SCfhzZ-RusXfw,5403
|
8
8
|
compressed_tensors/compressors/marlin_24.py,sha256=e7fGUyZbjUpA5VUMCPxqcYPGNiwoDKupHJaXWCoVKRw,9410
|
9
|
-
compressed_tensors/compressors/model_compressor.py,sha256=
|
9
|
+
compressed_tensors/compressors/model_compressor.py,sha256=3pMfGTTb8bN8PRNCFuH5k0RbP38r8GS_-cPgCkzL9vk,14355
|
10
10
|
compressed_tensors/compressors/naive_quantized.py,sha256=z3h3ca5xKCN69mahutxcbzdv-OysiaxaM8P-Qum6zUQ,4823
|
11
11
|
compressed_tensors/compressors/pack_quantized.py,sha256=27RVmJ2wg2dvCoawj407HSmKT3VPGJ6ujAMHlT26WlI,7571
|
12
12
|
compressed_tensors/compressors/sparse_bitmask.py,sha256=kiDwBlFV0sJGLcIdDYxIiuF64ccgwDfqq1hWRQThYDc,8647
|
@@ -28,7 +28,7 @@ compressed_tensors/quantization/lifecycle/compressed.py,sha256=Fj9n66IN0EWsOAkBH
|
|
28
28
|
compressed_tensors/quantization/lifecycle/forward.py,sha256=eLup6QDRUUp_Ozcas7RDRLIXBWjFbxn5gWbcAIJEGlw,15715
|
29
29
|
compressed_tensors/quantization/lifecycle/frozen.py,sha256=NiJw7NP7pcT6idWFa8vksgiLoT8oQ975e57S4QfD2QQ,1874
|
30
30
|
compressed_tensors/quantization/lifecycle/helpers.py,sha256=TmLY_G5VP_Fg2Ywio_dxoHRTxOKZdT7_aG5S9WtD4zI,2424
|
31
|
-
compressed_tensors/quantization/lifecycle/initialize.py,sha256=
|
31
|
+
compressed_tensors/quantization/lifecycle/initialize.py,sha256=HAtSm7vKOZ3kGZuWe2B8LsmfC5B5vIKlc0V8C4rAF4Y,8819
|
32
32
|
compressed_tensors/quantization/observers/__init__.py,sha256=4Sa7rqi5RB_S5bPO8KmncETiqDsoMBhwP37arlQym8s,764
|
33
33
|
compressed_tensors/quantization/observers/base.py,sha256=5ovQicWPYHjIxr6-EkQ4lgOX0PpI9g23iSzKpxjM1Zg,8420
|
34
34
|
compressed_tensors/quantization/observers/helpers.py,sha256=s_A23Qa_BLfOdHJCN5bm-qPWkhjjj_RIVrhSp1Y9Dtk,4211
|
@@ -46,8 +46,8 @@ compressed_tensors/utils/permutations_24.py,sha256=kx6fsfDHebx94zsSzhXGyCyuC9sVy
|
|
46
46
|
compressed_tensors/utils/permute.py,sha256=V6tJLKo3Syccj-viv4F7ZKZgJeCB-hl-dK8RKI_kBwI,2355
|
47
47
|
compressed_tensors/utils/safetensors_load.py,sha256=m08ANVuTBxQdoa6LufDgcNJ7wCLDJolyZljB8VEybAU,8578
|
48
48
|
compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
|
49
|
-
compressed_tensors_nightly-0.6.0.
|
50
|
-
compressed_tensors_nightly-0.6.0.
|
51
|
-
compressed_tensors_nightly-0.6.0.
|
52
|
-
compressed_tensors_nightly-0.6.0.
|
53
|
-
compressed_tensors_nightly-0.6.0.
|
49
|
+
compressed_tensors_nightly-0.6.0.20240928.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
|
50
|
+
compressed_tensors_nightly-0.6.0.20240928.dist-info/METADATA,sha256=vndAZXPsHUGFnoR1oLqalmP1tnMaAUx7QgXHPVrwarE,6799
|
51
|
+
compressed_tensors_nightly-0.6.0.20240928.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
|
52
|
+
compressed_tensors_nightly-0.6.0.20240928.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
|
53
|
+
compressed_tensors_nightly-0.6.0.20240928.dist-info/RECORD,,
|
File without changes
|
File without changes
|