compressed-tensors-nightly 0.4.0.20240617__py3-none-any.whl → 0.4.0.20240619__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/quantization/lifecycle/forward.py +8 -0
- compressed_tensors/quantization/quant_scheme.py +2 -2
- {compressed_tensors_nightly-0.4.0.20240617.dist-info → compressed_tensors_nightly-0.4.0.20240619.dist-info}/METADATA +1 -1
- {compressed_tensors_nightly-0.4.0.20240617.dist-info → compressed_tensors_nightly-0.4.0.20240619.dist-info}/RECORD +7 -7
- {compressed_tensors_nightly-0.4.0.20240617.dist-info → compressed_tensors_nightly-0.4.0.20240619.dist-info}/LICENSE +0 -0
- {compressed_tensors_nightly-0.4.0.20240617.dist-info → compressed_tensors_nightly-0.4.0.20240619.dist-info}/WHEEL +0 -0
- {compressed_tensors_nightly-0.4.0.20240617.dist-info → compressed_tensors_nightly-0.4.0.20240619.dist-info}/top_level.txt +0 -0
|
@@ -93,6 +93,14 @@ def dequantize(
|
|
|
93
93
|
:param args: quantization args used to quantize x_q
|
|
94
94
|
:return: dequantized float tensor
|
|
95
95
|
"""
|
|
96
|
+
# ensure all tensors are on the same device
|
|
97
|
+
# assumes that the target device is the input
|
|
98
|
+
# tensor's device
|
|
99
|
+
if x_q.device != scale.device:
|
|
100
|
+
scale = scale.to(x_q.device)
|
|
101
|
+
if x_q.device != zero_point.device:
|
|
102
|
+
zero_point = zero_point.to(x_q.device)
|
|
103
|
+
|
|
96
104
|
if args is None:
|
|
97
105
|
if scale.ndim == 0 or scale.ndim == 1:
|
|
98
106
|
args = QuantizationArgs(strategy=QuantizationStrategy.TENSOR)
|
|
@@ -108,10 +108,10 @@ def is_preset_scheme(name: str) -> bool:
|
|
|
108
108
|
|
|
109
109
|
|
|
110
110
|
W8A8 = dict(
|
|
111
|
-
weights=QuantizationArgs(), input_activations=QuantizationArgs(symmetric=
|
|
111
|
+
weights=QuantizationArgs(), input_activations=QuantizationArgs(symmetric=True)
|
|
112
112
|
)
|
|
113
113
|
|
|
114
|
-
W4A16 = dict(weights=QuantizationArgs(num_bits=4, symmetric=
|
|
114
|
+
W4A16 = dict(weights=QuantizationArgs(num_bits=4, symmetric=True))
|
|
115
115
|
|
|
116
116
|
PRESET_SCHEMES = {
|
|
117
117
|
"W8A8": W8A8,
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.1
|
|
2
2
|
Name: compressed-tensors-nightly
|
|
3
|
-
Version: 0.4.0.
|
|
3
|
+
Version: 0.4.0.20240619
|
|
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.
|
|
@@ -21,12 +21,12 @@ compressed_tensors/config/sparse_bitmask.py,sha256=pZUboRNZTu6NajGOQEFExoPknak5y
|
|
|
21
21
|
compressed_tensors/quantization/__init__.py,sha256=83J5bPB7PavN2TfCoW7_vEDhfYpm4TDrqYO9vdSQ5bk,760
|
|
22
22
|
compressed_tensors/quantization/quant_args.py,sha256=Z9Zu20ooAwEWlliAdUw1f1zwSrheuD6vqm3YXgJ1Lws,4388
|
|
23
23
|
compressed_tensors/quantization/quant_config.py,sha256=hL42sXp1wAZxyrkHarw7tAMRcwSVEr0MT3wmrmL3NhE,8285
|
|
24
|
-
compressed_tensors/quantization/quant_scheme.py,sha256
|
|
24
|
+
compressed_tensors/quantization/quant_scheme.py,sha256=3LX_UMS0uaYbBW1Zei-REuGXtNtFwAMk8lIj7lnHDP0,3569
|
|
25
25
|
compressed_tensors/quantization/lifecycle/__init__.py,sha256=ggRGWRqhCxCaTTDWRcgTVX3axnS2xV6rc5YvdzK7fSg,798
|
|
26
26
|
compressed_tensors/quantization/lifecycle/apply.py,sha256=aZrglJ5mR3Xaxwj51-1BVVB1JGVkKQEeHxGfBaVmsHI,8881
|
|
27
27
|
compressed_tensors/quantization/lifecycle/calibration.py,sha256=mLns4jlaWmBwOW8Jtlm5bMX-JET1AiZYUBO7qa-XuxI,1776
|
|
28
28
|
compressed_tensors/quantization/lifecycle/compressed.py,sha256=VreB10xPwgSLQQlTu20UCrFpRS--cA7-lx5s7nrPPrg,2247
|
|
29
|
-
compressed_tensors/quantization/lifecycle/forward.py,sha256=
|
|
29
|
+
compressed_tensors/quantization/lifecycle/forward.py,sha256=F4VQmCPXBpjWsDnlEErJ2eTWK54Ir9S1Kn2frlkBr9A,11401
|
|
30
30
|
compressed_tensors/quantization/lifecycle/frozen.py,sha256=h1XYt89MouBTf3jTYLG_6OdFxIu5q2N8tPjsy6J4E6Y,1726
|
|
31
31
|
compressed_tensors/quantization/lifecycle/initialize.py,sha256=9xgPzHejQUO_AkZcc_SH5kqFeieG-9uo0fMRYV51i7Y,4577
|
|
32
32
|
compressed_tensors/quantization/observers/__init__.py,sha256=DNH31NQYrIBBcmHsMyFA6whh4pbRsLwuNa6L8AeXaGc,745
|
|
@@ -41,8 +41,8 @@ compressed_tensors/registry/registry.py,sha256=fxjOjh2wklCvJhQxwofdy-zV8q7MkQ85S
|
|
|
41
41
|
compressed_tensors/utils/__init__.py,sha256=5DrYjoZbaEvSkJcC-GRSbM_RBHVF4tG9gMd3zsJnjLw,665
|
|
42
42
|
compressed_tensors/utils/helpers.py,sha256=5ull5yFT31M2zVxKeFvpvvlvX5f1Sk1LGuj_wrfZWCY,2267
|
|
43
43
|
compressed_tensors/utils/safetensors_load.py,sha256=0MheXwx1jeY12PeISppiSIZHs6rmN2YddwPpFb9V67I,8527
|
|
44
|
-
compressed_tensors_nightly-0.4.0.
|
|
45
|
-
compressed_tensors_nightly-0.4.0.
|
|
46
|
-
compressed_tensors_nightly-0.4.0.
|
|
47
|
-
compressed_tensors_nightly-0.4.0.
|
|
48
|
-
compressed_tensors_nightly-0.4.0.
|
|
44
|
+
compressed_tensors_nightly-0.4.0.20240619.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
|
|
45
|
+
compressed_tensors_nightly-0.4.0.20240619.dist-info/METADATA,sha256=xJpJU2rjediV-76_G-EwExDIQ_qTDgq2L3Gci1vpFdg,5668
|
|
46
|
+
compressed_tensors_nightly-0.4.0.20240619.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
|
|
47
|
+
compressed_tensors_nightly-0.4.0.20240619.dist-info/top_level.txt,sha256=w2i-GyPs2s1UwVxvutSvN_lM22SXC2hQFBmoMcPnV7Y,19
|
|
48
|
+
compressed_tensors_nightly-0.4.0.20240619.dist-info/RECORD,,
|
|
File without changes
|
|
File without changes
|