compressed-tensors-nightly 0.9.2.20250227__tar.gz → 0.9.2.20250301__tar.gz
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_nightly-0.9.2.20250227/src/compressed_tensors_nightly.egg-info → compressed_tensors_nightly-0.9.2.20250301}/PKG-INFO +1 -1
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/linear/compressed_linear.py +2 -1
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/lifecycle/initialize.py +2 -3
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301/src/compressed_tensors_nightly.egg-info}/PKG-INFO +1 -1
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/LICENSE +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/README.md +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/pyproject.toml +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/setup.cfg +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/setup.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/base.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/base.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/helpers.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/model_compressors/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/model_compressors/model_compressor.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/quantized_compressors/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/quantized_compressors/base.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/quantized_compressors/naive_quantized.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/quantized_compressors/pack_quantized.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/sparse_compressors/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/sparse_compressors/base.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/sparse_compressors/dense.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/sparse_compressors/sparse_24_bitmask.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/sparse_compressors/sparse_bitmask.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/sparse_quantized_compressors/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/compressors/sparse_quantized_compressors/marlin_24.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/config/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/config/base.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/config/dense.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/config/sparse_24_bitmask.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/config/sparse_bitmask.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/linear/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/lifecycle/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/lifecycle/apply.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/lifecycle/compressed.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/lifecycle/forward.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/lifecycle/helpers.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/quant_args.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/quant_config.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/quant_scheme.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/utils/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/quantization/utils/helpers.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/registry/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/registry/registry.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/utils/__init__.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/utils/helpers.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/utils/offload.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/utils/permutations_24.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/utils/permute.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/utils/safetensors_load.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/utils/semi_structured_conversions.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors/version.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors_nightly.egg-info/SOURCES.txt +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors_nightly.egg-info/dependency_links.txt +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors_nightly.egg-info/requires.txt +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/src/compressed_tensors_nightly.egg-info/top_level.txt +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/tests/test_registry.py +0 -0
- {compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/tests/testing_utils.py +0 -0
@@ -1,6 +1,6 @@
|
|
1
1
|
Metadata-Version: 2.2
|
2
2
|
Name: compressed-tensors-nightly
|
3
|
-
Version: 0.9.2.
|
3
|
+
Version: 0.9.2.20250301
|
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.
|
@@ -21,6 +21,7 @@ from compressed_tensors.quantization import (
|
|
21
21
|
QuantizationStatus,
|
22
22
|
initialize_module_for_quantization,
|
23
23
|
)
|
24
|
+
from compressed_tensors.utils import register_offload_parameter
|
24
25
|
from torch import Tensor
|
25
26
|
from torch.nn import Parameter
|
26
27
|
from torch.nn.functional import linear
|
@@ -68,7 +69,7 @@ class CompressedLinear(Linear):
|
|
68
69
|
param = Parameter(
|
69
70
|
torch.empty(shape, device=device, dtype=dtype), requires_grad=False
|
70
71
|
)
|
71
|
-
module
|
72
|
+
register_offload_parameter(module, name, param)
|
72
73
|
|
73
74
|
# mark module as compressed
|
74
75
|
module.quantization_status = QuantizationStatus.COMPRESSED
|
@@ -203,11 +203,10 @@ def _initialize_attn_scales(module: Module) -> None:
|
|
203
203
|
torch.empty(expected_shape, dtype=scale_dtype, device=device),
|
204
204
|
requires_grad=False,
|
205
205
|
)
|
206
|
-
|
207
|
-
module.register_parameter(KVCacheScaleType.KEY.value, init_scale)
|
206
|
+
register_offload_parameter(module, KVCacheScaleType.KEY.value, init_scale)
|
208
207
|
|
209
208
|
init_scale = Parameter(
|
210
209
|
torch.empty(expected_shape, dtype=scale_dtype, device=device),
|
211
210
|
requires_grad=False,
|
212
211
|
)
|
213
|
-
module
|
212
|
+
register_offload_parameter(module, KVCacheScaleType.VALUE.value, init_scale)
|
@@ -1,6 +1,6 @@
|
|
1
1
|
Metadata-Version: 2.2
|
2
2
|
Name: compressed-tensors-nightly
|
3
|
-
Version: 0.9.2.
|
3
|
+
Version: 0.9.2.20250301
|
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.
|
{compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/LICENSE
RENAMED
File without changes
|
{compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/README.md
RENAMED
File without changes
|
File without changes
|
{compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/setup.cfg
RENAMED
File without changes
|
{compressed_tensors_nightly-0.9.2.20250227 → compressed_tensors_nightly-0.9.2.20250301}/setup.py
RENAMED
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|