compressed-tensors-nightly 0.8.0.20241125__py3-none-any.whl → 0.8.0.20241127__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/quantized_compressors/naive_quantized.py +4 -2
- compressed_tensors/compressors/quantized_compressors/pack_quantized.py +2 -0
- compressed_tensors/quantization/quant_scheme.py +16 -2
- {compressed_tensors_nightly-0.8.0.20241125.dist-info → compressed_tensors_nightly-0.8.0.20241127.dist-info}/METADATA +1 -1
- {compressed_tensors_nightly-0.8.0.20241125.dist-info → compressed_tensors_nightly-0.8.0.20241127.dist-info}/RECORD +8 -8
- {compressed_tensors_nightly-0.8.0.20241125.dist-info → compressed_tensors_nightly-0.8.0.20241127.dist-info}/LICENSE +0 -0
- {compressed_tensors_nightly-0.8.0.20241125.dist-info → compressed_tensors_nightly-0.8.0.20241127.dist-info}/WHEEL +0 -0
- {compressed_tensors_nightly-0.8.0.20241125.dist-info → compressed_tensors_nightly-0.8.0.20241127.dist-info}/top_level.txt +0 -0
@@ -93,9 +93,11 @@ class NaiveQuantizationCompressor(BaseQuantizationCompressor):
|
|
93
93
|
args=quantization_args,
|
94
94
|
dtype=quantization_args.pytorch_dtype(),
|
95
95
|
)
|
96
|
+
else:
|
97
|
+
quantized_weight = weight
|
96
98
|
|
97
|
-
|
98
|
-
|
99
|
+
if device is not None:
|
100
|
+
quantized_weight = quantized_weight.to(device)
|
99
101
|
|
100
102
|
return {"weight": quantized_weight}
|
101
103
|
|
@@ -94,6 +94,8 @@ class PackedQuantizationCompressor(BaseQuantizationCompressor):
|
|
94
94
|
args=quantization_args,
|
95
95
|
dtype=torch.int8,
|
96
96
|
)
|
97
|
+
else:
|
98
|
+
quantized_weight = weight
|
97
99
|
|
98
100
|
packed_weight = pack_to_int32(quantized_weight, quantization_args.num_bits)
|
99
101
|
weight_shape = torch.tensor(weight.shape)
|
@@ -13,14 +13,14 @@
|
|
13
13
|
# limitations under the License.
|
14
14
|
|
15
15
|
from copy import deepcopy
|
16
|
-
from typing import List, Optional
|
16
|
+
from typing import Any, Dict, List, Optional
|
17
17
|
|
18
18
|
from compressed_tensors.quantization.quant_args import (
|
19
19
|
QuantizationArgs,
|
20
20
|
QuantizationStrategy,
|
21
21
|
QuantizationType,
|
22
22
|
)
|
23
|
-
from pydantic import BaseModel
|
23
|
+
from pydantic import BaseModel, model_validator
|
24
24
|
|
25
25
|
|
26
26
|
__all__ = [
|
@@ -47,6 +47,20 @@ class QuantizationScheme(BaseModel):
|
|
47
47
|
input_activations: Optional[QuantizationArgs] = None
|
48
48
|
output_activations: Optional[QuantizationArgs] = None
|
49
49
|
|
50
|
+
@model_validator(mode="after")
|
51
|
+
def validate_model_after(model: "QuantizationArgs") -> Dict[str, Any]:
|
52
|
+
inputs = model.input_activations
|
53
|
+
outputs = model.output_activations
|
54
|
+
|
55
|
+
if inputs is not None:
|
56
|
+
if inputs.actorder is not None:
|
57
|
+
raise ValueError("Cannot apply actorder to input activations")
|
58
|
+
|
59
|
+
if outputs is not None:
|
60
|
+
if outputs.actorder is not None:
|
61
|
+
raise ValueError("Cannot apply actorder to output activations")
|
62
|
+
|
63
|
+
return model
|
50
64
|
|
51
65
|
"""
|
52
66
|
Pre-Set Quantization Scheme Args
|
@@ -1,6 +1,6 @@
|
|
1
1
|
Metadata-Version: 2.1
|
2
2
|
Name: compressed-tensors-nightly
|
3
|
-
Version: 0.8.0.
|
3
|
+
Version: 0.8.0.20241127
|
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.
|
@@ -8,8 +8,8 @@ compressed_tensors/compressors/model_compressors/__init__.py,sha256=5RGGPFu4YqEt
|
|
8
8
|
compressed_tensors/compressors/model_compressors/model_compressor.py,sha256=sxh1TvW1Bp9YJE41hW0XZfd0kYYB85nhJvBLVRTDcV0,15886
|
9
9
|
compressed_tensors/compressors/quantized_compressors/__init__.py,sha256=09UJq68Pht6Bf-4iP9xYl3tetKsncNPHD8IAGbePsr4,714
|
10
10
|
compressed_tensors/compressors/quantized_compressors/base.py,sha256=K1KOnS6Y8nUA1-HN7VhyfsDc01nilW0WfXMUhuD-l8w,5954
|
11
|
-
compressed_tensors/compressors/quantized_compressors/naive_quantized.py,sha256=
|
12
|
-
compressed_tensors/compressors/quantized_compressors/pack_quantized.py,sha256=
|
11
|
+
compressed_tensors/compressors/quantized_compressors/naive_quantized.py,sha256=MMUya3Iwarm0BkeYXqKTUnEDPiBw98GKF09QiNST45k,4960
|
12
|
+
compressed_tensors/compressors/quantized_compressors/pack_quantized.py,sha256=1CLwvBlu4AtGkuo3IisD1-rQzwLiA6hE1bCc-pF_XGo,7758
|
13
13
|
compressed_tensors/compressors/sparse_compressors/__init__.py,sha256=i2TESH27l7KXeOhJ6hShIoI904XX96l-cRQiMR6MAaU,704
|
14
14
|
compressed_tensors/compressors/sparse_compressors/base.py,sha256=Ua4rUSGyucEs-YJI5z3oIUF-zqQLrFsQ9f-qKasEdUM,4410
|
15
15
|
compressed_tensors/compressors/sparse_compressors/dense.py,sha256=lSKNWRx6H7aUqaJj1j4qbXk8Gkm1UohbnvW1Rvq6Ra4,1284
|
@@ -25,7 +25,7 @@ compressed_tensors/linear/compressed_linear.py,sha256=MJa-UfoKhIkdUWRD1shrXXri2c
|
|
25
25
|
compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
|
26
26
|
compressed_tensors/quantization/quant_args.py,sha256=osjNwCSB6tcyH9Qeg5sHEiB-bHyi3XJ8TzkGVJuGTc4,8711
|
27
27
|
compressed_tensors/quantization/quant_config.py,sha256=K6kOZ6LDXpFlqsVzR4NEATV6y6Ea83rJWnNyVlvw-pI,10379
|
28
|
-
compressed_tensors/quantization/quant_scheme.py,sha256=
|
28
|
+
compressed_tensors/quantization/quant_scheme.py,sha256=w1RbYA3TzbCt9Us21hPvjxZW0jg2vVF1h9mWYUFhJpo,6009
|
29
29
|
compressed_tensors/quantization/lifecycle/__init__.py,sha256=_uItzFWusyV74Zco_pHLOTdE9a83cL-R-ZdyQrBkIyw,772
|
30
30
|
compressed_tensors/quantization/lifecycle/apply.py,sha256=jCUSgeOBtagE5IhgIbyYMZ4kv8Rm20VGJ4IxXZ5HAnw,15066
|
31
31
|
compressed_tensors/quantization/lifecycle/compressed.py,sha256=Fj9n66IN0EWsOAkBHg3O0GlOQpxstqjCcs0ttzMXrJ0,2296
|
@@ -43,8 +43,8 @@ compressed_tensors/utils/permutations_24.py,sha256=kx6fsfDHebx94zsSzhXGyCyuC9sVy
|
|
43
43
|
compressed_tensors/utils/permute.py,sha256=V6tJLKo3Syccj-viv4F7ZKZgJeCB-hl-dK8RKI_kBwI,2355
|
44
44
|
compressed_tensors/utils/safetensors_load.py,sha256=m08ANVuTBxQdoa6LufDgcNJ7wCLDJolyZljB8VEybAU,8578
|
45
45
|
compressed_tensors/utils/semi_structured_conversions.py,sha256=XKNffPum54kPASgqKzgKvyeqWPAkair2XEQXjkp7ho8,13489
|
46
|
-
compressed_tensors_nightly-0.8.0.
|
47
|
-
compressed_tensors_nightly-0.8.0.
|
48
|
-
compressed_tensors_nightly-0.8.0.
|
49
|
-
compressed_tensors_nightly-0.8.0.
|
50
|
-
compressed_tensors_nightly-0.8.0.
|
46
|
+
compressed_tensors_nightly-0.8.0.20241127.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
|
47
|
+
compressed_tensors_nightly-0.8.0.20241127.dist-info/METADATA,sha256=JxFY_KbCnr3oLtEZrQLaEtR4HXUhSuUoYKKeyMZmX9U,6799
|
48
|
+
compressed_tensors_nightly-0.8.0.20241127.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
|
49
|
+
compressed_tensors_nightly-0.8.0.20241127.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
|
50
|
+
compressed_tensors_nightly-0.8.0.20241127.dist-info/RECORD,,
|
File without changes
|
File without changes
|