torchquery 2.1.0__tar.gz → 2.1.2__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.
- torchquery-2.1.2/PKG-INFO +68 -0
- torchquery-2.1.2/README.md +45 -0
- {torchquery-2.1.0 → torchquery-2.1.2}/pyproject.toml +2 -2
- {torchquery-2.1.0 → torchquery-2.1.2}/setup.py +1 -1
- torchquery-2.1.2/torchquery.egg-info/PKG-INFO +68 -0
- torchquery-2.1.0/PKG-INFO +0 -37
- torchquery-2.1.0/README.md +0 -14
- torchquery-2.1.0/torchquery.egg-info/PKG-INFO +0 -37
- {torchquery-2.1.0 → torchquery-2.1.2}/LICENSE +0 -0
- {torchquery-2.1.0 → torchquery-2.1.2}/setup.cfg +0 -0
- {torchquery-2.1.0 → torchquery-2.1.2}/torchquery.egg-info/SOURCES.txt +0 -0
- {torchquery-2.1.0 → torchquery-2.1.2}/torchquery.egg-info/dependency_links.txt +0 -0
- {torchquery-2.1.0 → torchquery-2.1.2}/torchquery.egg-info/requires.txt +0 -0
- {torchquery-2.1.0 → torchquery-2.1.2}/torchquery.egg-info/top_level.txt +0 -0
|
@@ -0,0 +1,68 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: torchquery
|
|
3
|
+
Version: 2.1.2
|
|
4
|
+
Summary: High-performance SDC detection and neural healing for billion-scale tensors.
|
|
5
|
+
Home-page: https://github.com/powerofaisinstudy-debug/torchquery
|
|
6
|
+
Author: Sundaram Gupta
|
|
7
|
+
Author-email: your-email@example.com
|
|
8
|
+
Project-URL: Homepage, https://github.com/powerofaisinstudy-debug/torchquery
|
|
9
|
+
Project-URL: Bug Tracker, https://github.com/powerofaisinstudy-debug/torchquery/issues
|
|
10
|
+
Classifier: Programming Language :: Python :: 3
|
|
11
|
+
Classifier: License :: OSI Approved :: MIT License
|
|
12
|
+
Classifier: Operating System :: OS Independent
|
|
13
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
14
|
+
Requires-Python: >=3.7
|
|
15
|
+
Description-Content-Type: text/markdown
|
|
16
|
+
License-File: LICENSE
|
|
17
|
+
Requires-Dist: torch>=1.9.0
|
|
18
|
+
Requires-Dist: numpy
|
|
19
|
+
Dynamic: author-email
|
|
20
|
+
Dynamic: home-page
|
|
21
|
+
Dynamic: license-file
|
|
22
|
+
Dynamic: requires-python
|
|
23
|
+
|
|
24
|
+
# TorchQuery 🛡️
|
|
25
|
+
|
|
26
|
+
<p align="center">
|
|
27
|
+
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/tch.png" width="600" alt="TorchQuery Logo">
|
|
28
|
+
</p>
|
|
29
|
+
|
|
30
|
+
---
|
|
31
|
+
|
|
32
|
+
[](https://pypi.org/project/torchquery/)
|
|
33
|
+
[](https://opensource.org/licenses/MIT)
|
|
34
|
+
|
|
35
|
+
**TorchQuery** is a high-performance reliability engine for PyTorch. It provides a "Neural Shield" against **Silent Data Corruption (SDC)**, hardware bit-flips, and numerical instability in massive Deep Learning models.
|
|
36
|
+
|
|
37
|
+
## 🚀 Key Features
|
|
38
|
+
|
|
39
|
+
* **Billion-Scale Protection:** Optimized streaming logic designed to handle tensors with $10^9$ elements without crashing.
|
|
40
|
+
* **Neural Healing:** Automatically detects and repairs corrupted weights or neurons using statistical outlier detection ($\sigma$-clamping).
|
|
41
|
+
* **Distributed SyncBatch:** Cluster-aware protection using `All-Reduce` to ensure safety across multi-GPU and multi-server environments.
|
|
42
|
+
* **Zero-Invasive:** Simply wrap your existing tensors or model parameters; no architecture changes required.
|
|
43
|
+
|
|
44
|
+
---
|
|
45
|
+
|
|
46
|
+
## 📦 Installation
|
|
47
|
+
|
|
48
|
+
```bash
|
|
49
|
+
pip install torchquery
|
|
50
|
+
|
|
51
|
+
<p align="center">
|
|
52
|
+
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/chl.png" width="600">
|
|
53
|
+
</p>
|
|
54
|
+
|
|
55
|
+
### Visualizing Silent Data Corruption (SDC)
|
|
56
|
+
|
|
57
|
+
Hardware glitches—like cosmic rays or VRAM overclocks—can cause random bit-flips. These create massive statistical outliers or `NaNs` in your tensor data.
|
|
58
|
+
|
|
59
|
+
[Image Link to Image_5.png]
|
|
60
|
+
|
|
61
|
+
**TorchQuery** acts as a `Neural Shield` that sweeps your multidimensional arrays. It identifies values that can lead to exploding gradients (`3e38`) or numerical instability (`NaN`), "healing" them before they propagate.
|
|
62
|
+
|
|
63
|
+
**Pre-Sweep State:**
|
|
64
|
+
* `NaN` (Not a Number): Corrupts entire model during backpropagation.
|
|
65
|
+
* `3e38`: Causes exploding gradients, destroying training stability.
|
|
66
|
+
|
|
67
|
+
**Post-Sweep State:**
|
|
68
|
+
* Invalid data is removed, leaving behind validated tensor values.
|
|
@@ -0,0 +1,45 @@
|
|
|
1
|
+
# TorchQuery 🛡️
|
|
2
|
+
|
|
3
|
+
<p align="center">
|
|
4
|
+
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/tch.png" width="600" alt="TorchQuery Logo">
|
|
5
|
+
</p>
|
|
6
|
+
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
[](https://pypi.org/project/torchquery/)
|
|
10
|
+
[](https://opensource.org/licenses/MIT)
|
|
11
|
+
|
|
12
|
+
**TorchQuery** is a high-performance reliability engine for PyTorch. It provides a "Neural Shield" against **Silent Data Corruption (SDC)**, hardware bit-flips, and numerical instability in massive Deep Learning models.
|
|
13
|
+
|
|
14
|
+
## 🚀 Key Features
|
|
15
|
+
|
|
16
|
+
* **Billion-Scale Protection:** Optimized streaming logic designed to handle tensors with $10^9$ elements without crashing.
|
|
17
|
+
* **Neural Healing:** Automatically detects and repairs corrupted weights or neurons using statistical outlier detection ($\sigma$-clamping).
|
|
18
|
+
* **Distributed SyncBatch:** Cluster-aware protection using `All-Reduce` to ensure safety across multi-GPU and multi-server environments.
|
|
19
|
+
* **Zero-Invasive:** Simply wrap your existing tensors or model parameters; no architecture changes required.
|
|
20
|
+
|
|
21
|
+
---
|
|
22
|
+
|
|
23
|
+
## 📦 Installation
|
|
24
|
+
|
|
25
|
+
```bash
|
|
26
|
+
pip install torchquery
|
|
27
|
+
|
|
28
|
+
<p align="center">
|
|
29
|
+
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/chl.png" width="600">
|
|
30
|
+
</p>
|
|
31
|
+
|
|
32
|
+
### Visualizing Silent Data Corruption (SDC)
|
|
33
|
+
|
|
34
|
+
Hardware glitches—like cosmic rays or VRAM overclocks—can cause random bit-flips. These create massive statistical outliers or `NaNs` in your tensor data.
|
|
35
|
+
|
|
36
|
+
[Image Link to Image_5.png]
|
|
37
|
+
|
|
38
|
+
**TorchQuery** acts as a `Neural Shield` that sweeps your multidimensional arrays. It identifies values that can lead to exploding gradients (`3e38`) or numerical instability (`NaN`), "healing" them before they propagate.
|
|
39
|
+
|
|
40
|
+
**Pre-Sweep State:**
|
|
41
|
+
* `NaN` (Not a Number): Corrupts entire model during backpropagation.
|
|
42
|
+
* `3e38`: Causes exploding gradients, destroying training stability.
|
|
43
|
+
|
|
44
|
+
**Post-Sweep State:**
|
|
45
|
+
* Invalid data is removed, leaving behind validated tensor values.
|
|
@@ -4,9 +4,9 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "torchquery"
|
|
7
|
-
version = "2.1.
|
|
7
|
+
version = "2.1.2"
|
|
8
8
|
authors = [
|
|
9
|
-
{ name="Sundaram Gupta"
|
|
9
|
+
{ name="Sundaram Gupta"},
|
|
10
10
|
]
|
|
11
11
|
description = "High-performance SDC detection and neural healing for billion-scale tensors."
|
|
12
12
|
readme = "README.md"
|
|
@@ -8,7 +8,7 @@ with open("README.md", "r", encoding="utf-8") as fh:
|
|
|
8
8
|
|
|
9
9
|
setup(
|
|
10
10
|
name="torchquery",
|
|
11
|
-
version="2.1.
|
|
11
|
+
version="2.1.2",
|
|
12
12
|
author="Sundaram Gupta",
|
|
13
13
|
author_email="your-email@example.com", # Update this with your email
|
|
14
14
|
description="A high-performance SDC detection and neural healing engine for billion-scale tensors.",
|
|
@@ -0,0 +1,68 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: torchquery
|
|
3
|
+
Version: 2.1.2
|
|
4
|
+
Summary: High-performance SDC detection and neural healing for billion-scale tensors.
|
|
5
|
+
Home-page: https://github.com/powerofaisinstudy-debug/torchquery
|
|
6
|
+
Author: Sundaram Gupta
|
|
7
|
+
Author-email: your-email@example.com
|
|
8
|
+
Project-URL: Homepage, https://github.com/powerofaisinstudy-debug/torchquery
|
|
9
|
+
Project-URL: Bug Tracker, https://github.com/powerofaisinstudy-debug/torchquery/issues
|
|
10
|
+
Classifier: Programming Language :: Python :: 3
|
|
11
|
+
Classifier: License :: OSI Approved :: MIT License
|
|
12
|
+
Classifier: Operating System :: OS Independent
|
|
13
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
14
|
+
Requires-Python: >=3.7
|
|
15
|
+
Description-Content-Type: text/markdown
|
|
16
|
+
License-File: LICENSE
|
|
17
|
+
Requires-Dist: torch>=1.9.0
|
|
18
|
+
Requires-Dist: numpy
|
|
19
|
+
Dynamic: author-email
|
|
20
|
+
Dynamic: home-page
|
|
21
|
+
Dynamic: license-file
|
|
22
|
+
Dynamic: requires-python
|
|
23
|
+
|
|
24
|
+
# TorchQuery 🛡️
|
|
25
|
+
|
|
26
|
+
<p align="center">
|
|
27
|
+
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/tch.png" width="600" alt="TorchQuery Logo">
|
|
28
|
+
</p>
|
|
29
|
+
|
|
30
|
+
---
|
|
31
|
+
|
|
32
|
+
[](https://pypi.org/project/torchquery/)
|
|
33
|
+
[](https://opensource.org/licenses/MIT)
|
|
34
|
+
|
|
35
|
+
**TorchQuery** is a high-performance reliability engine for PyTorch. It provides a "Neural Shield" against **Silent Data Corruption (SDC)**, hardware bit-flips, and numerical instability in massive Deep Learning models.
|
|
36
|
+
|
|
37
|
+
## 🚀 Key Features
|
|
38
|
+
|
|
39
|
+
* **Billion-Scale Protection:** Optimized streaming logic designed to handle tensors with $10^9$ elements without crashing.
|
|
40
|
+
* **Neural Healing:** Automatically detects and repairs corrupted weights or neurons using statistical outlier detection ($\sigma$-clamping).
|
|
41
|
+
* **Distributed SyncBatch:** Cluster-aware protection using `All-Reduce` to ensure safety across multi-GPU and multi-server environments.
|
|
42
|
+
* **Zero-Invasive:** Simply wrap your existing tensors or model parameters; no architecture changes required.
|
|
43
|
+
|
|
44
|
+
---
|
|
45
|
+
|
|
46
|
+
## 📦 Installation
|
|
47
|
+
|
|
48
|
+
```bash
|
|
49
|
+
pip install torchquery
|
|
50
|
+
|
|
51
|
+
<p align="center">
|
|
52
|
+
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/chl.png" width="600">
|
|
53
|
+
</p>
|
|
54
|
+
|
|
55
|
+
### Visualizing Silent Data Corruption (SDC)
|
|
56
|
+
|
|
57
|
+
Hardware glitches—like cosmic rays or VRAM overclocks—can cause random bit-flips. These create massive statistical outliers or `NaNs` in your tensor data.
|
|
58
|
+
|
|
59
|
+
[Image Link to Image_5.png]
|
|
60
|
+
|
|
61
|
+
**TorchQuery** acts as a `Neural Shield` that sweeps your multidimensional arrays. It identifies values that can lead to exploding gradients (`3e38`) or numerical instability (`NaN`), "healing" them before they propagate.
|
|
62
|
+
|
|
63
|
+
**Pre-Sweep State:**
|
|
64
|
+
* `NaN` (Not a Number): Corrupts entire model during backpropagation.
|
|
65
|
+
* `3e38`: Causes exploding gradients, destroying training stability.
|
|
66
|
+
|
|
67
|
+
**Post-Sweep State:**
|
|
68
|
+
* Invalid data is removed, leaving behind validated tensor values.
|
torchquery-2.1.0/PKG-INFO
DELETED
|
@@ -1,37 +0,0 @@
|
|
|
1
|
-
Metadata-Version: 2.4
|
|
2
|
-
Name: torchquery
|
|
3
|
-
Version: 2.1.0
|
|
4
|
-
Summary: High-performance SDC detection and neural healing for billion-scale tensors.
|
|
5
|
-
Home-page: https://github.com/powerofaisinstudy-debug/torchquery
|
|
6
|
-
Author: Sundaram Gupta
|
|
7
|
-
Author-email: Sundaram Gupta <your-email@example.com>
|
|
8
|
-
Project-URL: Homepage, https://github.com/powerofaisinstudy-debug/torchquery
|
|
9
|
-
Project-URL: Bug Tracker, https://github.com/powerofaisinstudy-debug/torchquery/issues
|
|
10
|
-
Classifier: Programming Language :: Python :: 3
|
|
11
|
-
Classifier: License :: OSI Approved :: MIT License
|
|
12
|
-
Classifier: Operating System :: OS Independent
|
|
13
|
-
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
14
|
-
Requires-Python: >=3.7
|
|
15
|
-
Description-Content-Type: text/markdown
|
|
16
|
-
License-File: LICENSE
|
|
17
|
-
Requires-Dist: torch>=1.9.0
|
|
18
|
-
Requires-Dist: numpy
|
|
19
|
-
Dynamic: author
|
|
20
|
-
Dynamic: home-page
|
|
21
|
-
Dynamic: license-file
|
|
22
|
-
Dynamic: requires-python
|
|
23
|
-
|
|
24
|
-
<p align="center">
|
|
25
|
-
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/tch.png" width="600" alt="TorchQuery Logo">
|
|
26
|
-
</p>" width="600" alt="TorchQuery Logo">
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
# TorchQuery 🚀
|
|
30
|
-
|
|
31
|
-
**TorchQuery** is a high-performance, vectorized engine built on top of PyTorch. It is designed for **Billion-Scale Neural Healing** and advanced tensor manipulation without using Python loops.
|
|
32
|
-
|
|
33
|
-
Created by **Sundaram Gupta**.
|
|
34
|
-
|
|
35
|
-
## Installation
|
|
36
|
-
```bash
|
|
37
|
-
pip install torchquery
|
torchquery-2.1.0/README.md
DELETED
|
@@ -1,14 +0,0 @@
|
|
|
1
|
-
<p align="center">
|
|
2
|
-
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/tch.png" width="600" alt="TorchQuery Logo">
|
|
3
|
-
</p>" width="600" alt="TorchQuery Logo">
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
# TorchQuery 🚀
|
|
7
|
-
|
|
8
|
-
**TorchQuery** is a high-performance, vectorized engine built on top of PyTorch. It is designed for **Billion-Scale Neural Healing** and advanced tensor manipulation without using Python loops.
|
|
9
|
-
|
|
10
|
-
Created by **Sundaram Gupta**.
|
|
11
|
-
|
|
12
|
-
## Installation
|
|
13
|
-
```bash
|
|
14
|
-
pip install torchquery
|
|
@@ -1,37 +0,0 @@
|
|
|
1
|
-
Metadata-Version: 2.4
|
|
2
|
-
Name: torchquery
|
|
3
|
-
Version: 2.1.0
|
|
4
|
-
Summary: High-performance SDC detection and neural healing for billion-scale tensors.
|
|
5
|
-
Home-page: https://github.com/powerofaisinstudy-debug/torchquery
|
|
6
|
-
Author: Sundaram Gupta
|
|
7
|
-
Author-email: Sundaram Gupta <your-email@example.com>
|
|
8
|
-
Project-URL: Homepage, https://github.com/powerofaisinstudy-debug/torchquery
|
|
9
|
-
Project-URL: Bug Tracker, https://github.com/powerofaisinstudy-debug/torchquery/issues
|
|
10
|
-
Classifier: Programming Language :: Python :: 3
|
|
11
|
-
Classifier: License :: OSI Approved :: MIT License
|
|
12
|
-
Classifier: Operating System :: OS Independent
|
|
13
|
-
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
14
|
-
Requires-Python: >=3.7
|
|
15
|
-
Description-Content-Type: text/markdown
|
|
16
|
-
License-File: LICENSE
|
|
17
|
-
Requires-Dist: torch>=1.9.0
|
|
18
|
-
Requires-Dist: numpy
|
|
19
|
-
Dynamic: author
|
|
20
|
-
Dynamic: home-page
|
|
21
|
-
Dynamic: license-file
|
|
22
|
-
Dynamic: requires-python
|
|
23
|
-
|
|
24
|
-
<p align="center">
|
|
25
|
-
<img src="https://raw.githubusercontent.com/powerofaisinstudy-debug/torchquery/main/tch.png" width="600" alt="TorchQuery Logo">
|
|
26
|
-
</p>" width="600" alt="TorchQuery Logo">
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
# TorchQuery 🚀
|
|
30
|
-
|
|
31
|
-
**TorchQuery** is a high-performance, vectorized engine built on top of PyTorch. It is designed for **Billion-Scale Neural Healing** and advanced tensor manipulation without using Python loops.
|
|
32
|
-
|
|
33
|
-
Created by **Sundaram Gupta**.
|
|
34
|
-
|
|
35
|
-
## Installation
|
|
36
|
-
```bash
|
|
37
|
-
pip install torchquery
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|