# paGating: Parameterized Activation Gating Framework
[](https://pytorch.org/)
[](https://opensource.org/licenses/Apache-2.0)
[](https://python.org/)
[](./tests/)
[](https://github.com/guglxni/paGating)
> **๐ Production-Ready Framework for Parameterized Activation Gating in Neural Networks**
A comprehensive, open-source framework that unifies gated activation functions through a single parameterization scheme. Featured in our IEEE TNNLS submission: *"paGating: A Parameterized Activation Gating Framework for Flexible and Efficient Neural Networks for GenAI"*.
## ๐ฏ Key Results
Our framework demonstrates significant improvements across multiple domains:
| Domain | Metric | Improvement | Hardware |
|--------|--------|-------------|----------|
| **Language Modeling** | WikiText-103 Eval Loss | **1.9% improvement** | GPT-2 Small |
| **Image Classification** | CIFAR-10 Accuracy | **+1.9 percentage points** | ResNet variants |
| **Hardware Efficiency** | Apple M4 Inference | **3.11ร speedup** | 15% memory reduction |
## ๐ Features
- **๐ฌ 7 Core Gating Units + Specialized Components**: paGLU, paGTU, paSwishU, paReGLU, paGELU, paMishU, paSiLU, paUnit (template), PaGRUCell
- **โก Production Ready**: ONNX and CoreML export pipelines for deployment
- **๐งช Comprehensive Testing**: 93% test coverage with continuous integration
- **๐ Benchmarking Tools**: Built-in performance analysis and visualization
- **๐ PyTorch Lightning**: Seamless integration with modern training workflows
- **๐ฑ Cross-Platform**: CPU, CUDA, MPS (Apple Silicon) support
- **๐๏ธ Flexible Alpha**: Fixed, learnable, or scheduled parameter control
## Project Structure
The project has been organized into the following structure:
```
paGating/
โโโ assets/ # Static assets
โ โโโ images/ # Image files
โ โโโ figures/ # Paper figures
โ โโโ plots/ # Plot outputs from experiments
โโโ benchmark_results/ # Results from various benchmarks
โ โโโ coreml/ # CoreML benchmark results
โ โโโ regression/ # Regression task results
โ โโโ transformer/ # Transformer model results
โโโ coreml_models/ # Exported CoreML models
โโโ datamodules/ # PyTorch Lightning data modules
โโโ docs/ # Documentation
โ โโโ paper/ # Research paper and references
โ โโโ results_summary.md # Summary of experiment results
โโโ experiments/ # Experiment configurations
โโโ lightning_modules/ # PyTorch Lightning modules
โโโ models/ # Model implementations
โโโ onnx_models/ # Exported ONNX models
โโโ paGating/ # Core package
โ โโโ __init__.py # Package exports
โ โโโ base.py # Base classes
โ โโโ paGLU.py # Gated Linear Unit implementation
โ โโโ paGTU.py # Gated Tanh Unit implementation
โ โโโ paSwishU.py # Swish Unit implementation
โ โโโ paReGLU.py # ReLU Gated Linear Unit implementation
โ โโโ paGELU.py # GELU Gated Unit implementation
โ โโโ paMishU.py # Mish Unit implementation
โ โโโ paSiLU.py # SiLU/Swish gating implementation
โ โโโ paUnit.py # Generic gating unit template
โ โโโ paGRU.py # Parameterized GRU cell
โโโ scripts/ # Utility scripts
โ โโโ benchmark/ # Benchmarking scripts
โ โโโ utilities/ # Utility scripts
โโโ src/ # Source code (application-specific)
โโโ tests/ # Test suite
โโโ requirements.txt # Project dependencies
โโโ README.md # This file
```
## Implemented Gating Units
| Unit | Description | Formula |
|------|-------------|---------|
| paGLU | Parameterized Gated Linear Unit | x * (ฮฑ * sigmoid(x) + (1-ฮฑ)) |
| paGTU | Parameterized Gated Tanh Unit | x * (ฮฑ * tanh(x) + (1-ฮฑ)) |
| paSwishU | Parameterized Swish Unit | x * (ฮฑ * sigmoid(x) + (1-ฮฑ) * x) |
| paReGLU | Parameterized ReLU Gated Linear Unit | x * (ฮฑ * ReLU(x) + (1-ฮฑ)) |
| paGELU | Parameterized Gated GELU | x * (ฮฑ * GELU(x) + (1-ฮฑ)) |
| paMishU | Parameterized Mish Unit | x * (ฮฑ * mish(x) + (1-ฮฑ)) |
| paSiLU | Parameterized SiLU/Swish gating | x * (ฮฑ * SiLU(x) + (1-ฮฑ) * x) |
| paUnit | Generic Template for Custom Units | x * (ฮฑ * custom_fn(x) + (1-ฮฑ)) |
| PaGRUCell | Parameterized GRU Cell | Specialized recurrent architecture |
## Installation
Clone the repository:
```bash
git clone https://github.com/guglxni/paGating.git
cd paGating
```
Install requirements:
```bash
pip install -r requirements.txt
```
Set up data directories and download datasets:
```bash
python scripts/download_data.py
```
> **Note**: This repository uses symlinks for large data files. See [docs/DATA_SETUP.md](docs/DATA_SETUP.md) for detailed setup instructions.
## Quick Start
### Using a paGating unit in your model
```python
import torch
from paGating import paGLU
# Create a layer with fixed alpha
gating_layer = paGLU(input_dim=512, output_dim=512, alpha=0.5)
# Or with learnable alpha
learnable_gating_layer = paGLU(input_dim=512, output_dim=512, learnable_alpha=True)
# Use in a model
x = torch.randn(32, 512) # batch_size, input_dim
output = gating_layer(x) # shape: (32, 512)
```
### Integration with PyTorch models
```python
import torch
import torch.nn as nn
from paGating import paGLU
class MyModel(nn.Module):
def __init__(self):
super().__init__()
self.fc1 = nn.Linear(784, 512)
self.gate = paGLU(512, 512, alpha=0.5) # paGating unit
self.fc2 = nn.Linear(512, 10)
def forward(self, x):
x = self.fc1(x)
x = self.gate(x)
x = self.fc2(x)
return x
```
## Experimenting with paGating
### Running Benchmarks
The framework includes tools for benchmarking different gating units:
```bash
python scripts/benchmark/benchmark_gateflow.py
```
This generates plots comparing the performance of different units.
### Running a Hyperparameter Sweep
To compare different units and alpha values:
```bash
python scripts/utilities/run_experiment_pipeline.py --experiment_name my_experiment --units paGLU paGTU paMishU --alpha_values 0.0 0.2 0.5 0.8 1.0
```
This will:
1. Run a hyperparameter sweep
2. Generate a leaderboard
3. Create visualizations
### Testing with Transformer Models
To test a gating unit in a transformer for sequence classification:
```bash
python experiments/test_transformer.py --unit paMishU --alpha 0.5 --epochs 20
```
## Export to CoreML
You can export trained models to CoreML format for deployment on Apple devices:
```bash
python scripts/coreml_export.py --unit paGLU --alpha 0.5
```
Test the exported model:
```bash
python tests/test_coreml_model.py --unit paGLU --alpha 0.5
```
## Results Summary
For detailed results and comparisons of different gating units, see [docs/results_summary.md](docs/results_summary.md).
## Creating Your Own Gating Unit
To create a custom gating unit:
1. Create a new file in the paGating directory (e.g., `paGating/paMyCustomU.py`)
2. Extend the `paGatingBase` class
3. Implement the required methods
4. Update `__init__.py` to expose your new unit
Example:
```python
from .base import paGatingBase
import torch
import torch.nn as nn
import torch.nn.functional as F
class paMyCustomU(paGatingBase):
"""
My custom parameterized activation gating unit.
"""
def __init__(self, input_dim, output_dim, alpha=0.5, learnable_alpha=False, alpha_init=None, bias=True):
super().__init__(
input_dim=input_dim,
output_dim=output_dim,
alpha=alpha,
learnable_alpha=learnable_alpha,
alpha_init=alpha_init,
bias=bias
)
def compute_gate_activation(self, x):
# Implement your custom activation function
return my_custom_activation(x)
def forward(self, x):
# Standard implementation, can be customized if needed
x = self.linear(x)
gates = self.compute_gate_activation(x)
return x * gates
```
Then update `__init__.py`:
```python
from .paMyCustomU import paMyCustomU
__all__ = [
# ... existing units
'paMyCustomU',
]
```
## License
This project is licensed under the Apache License 2.0 - see the [LICENSE](LICENSE) file for details.
**Commercial Use**: For commercial applications, please contact the authors for licensing arrangements.
## ๐ Research Paper
This framework is featured in our IEEE TNNLS submission:
**"paGating: A Parameterized Activation Gating Framework for Flexible and Efficient Neural Networks for GenAI"**
- **Authors**: Aaryan Guglani, Dr. Rajashree Shettar
- **Institution**: RV College of Engineering, Bengaluru
- **Status**: Under Review at IEEE Transactions on Neural Networks and Learning Systems
- **Reproducibility**: Complete reproduction guide available in [`docs/REPRODUCIBILITY.md`](docs/REPRODUCIBILITY.md)
## ๐ Documentation
- **[Reproducibility Guide](docs/REPRODUCIBILITY.md)**: Step-by-step instructions to reproduce all paper results
- **[Contributing Guide](CONTRIBUTING.md)**: How to contribute to the project
- **[API Documentation](docs/)**: Detailed API reference and examples
## ๐ Citation
If you use paGating in your research, please cite:
```bibtex
@article{guglani2025pagating,
title={paGating: A Parameterized Activation Gating Framework for Flexible and Efficient Neural Networks for GenAI},
author={Guglani, Aaryan and Shettar, Rajashree},
journal={IEEE Transactions on Neural Networks and Learning Systems},
year={2025},
note={Under Review},
url={https://github.com/guglxni/paGating}
}
```
## Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
## Interactive Dashboard
The project includes a Streamlit dashboard for visualizing experiment results:
```bash
# Install required packages if not already installed
pip install streamlit plotly pandas
# Run the dashboard with a specific results directory
streamlit run scripts/streamlit_dashboard.py -- --results_dir results/your_experiment_dir
# Or run the dashboard and select the results directory in the UI
streamlit run scripts/streamlit_dashboard.py
```
Dashboard features:
- Compare performance across different gating units
- Analyze the effect of different alpha values
- Explore the behavior of learnable alpha parameters
- View training curves and leaderboards
- Generate insights and recommendations
## Experiments
Run a hyperparameter sweep:
```bash
python scripts/utilities/run_experiment_pipeline.py
```
This will:
1. Run a sweep over different units and alpha values
2. Generate a leaderboard
3. Create visualizations
4. Run the analysis
## Research Paper
A detailed research paper describing the paGating framework, its implementation, and experimental results is available in the [docs/paper/](docs/paper/) directory.
Raw data
{
"_id": null,
"home_page": "https://github.com/guglxni/paGating",
"name": "paGating",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.8",
"maintainer_email": "Aaryan Guglani <aaryanguglani.cs21@rvce.edu.in>",
"keywords": "deep-learning, neural-networks, activation-functions, pytorch, gating, transformer, machine-learning",
"author": "Aaryan Guglani",
"author_email": "Aaryan Guglani <aaryanguglani.cs21@rvce.edu.in>",
"download_url": "https://files.pythonhosted.org/packages/d8/1a/6186a14081ffe47c3084b3314f2a4a9b4caef3f193cb562dd319fc2e32eb/pagating-0.1.0.tar.gz",
"platform": null,
"description": "# paGating: Parameterized Activation Gating Framework\n\n[](https://pytorch.org/)\n[](https://opensource.org/licenses/Apache-2.0)\n[](https://python.org/)\n[](./tests/)\n[](https://github.com/guglxni/paGating)\n\n> **\ud83d\ude80 Production-Ready Framework for Parameterized Activation Gating in Neural Networks**\n\nA comprehensive, open-source framework that unifies gated activation functions through a single parameterization scheme. Featured in our IEEE TNNLS submission: *\"paGating: A Parameterized Activation Gating Framework for Flexible and Efficient Neural Networks for GenAI\"*.\n\n## \ud83c\udfaf Key Results\n\nOur framework demonstrates significant improvements across multiple domains:\n\n| Domain | Metric | Improvement | Hardware |\n|--------|--------|-------------|----------|\n| **Language Modeling** | WikiText-103 Eval Loss | **1.9% improvement** | GPT-2 Small |\n| **Image Classification** | CIFAR-10 Accuracy | **+1.9 percentage points** | ResNet variants |\n| **Hardware Efficiency** | Apple M4 Inference | **3.11\u00d7 speedup** | 15% memory reduction |\n\n## \ud83d\ude80 Features\n\n- **\ud83d\udd2c 7 Core Gating Units + Specialized Components**: paGLU, paGTU, paSwishU, paReGLU, paGELU, paMishU, paSiLU, paUnit (template), PaGRUCell\n- **\u26a1 Production Ready**: ONNX and CoreML export pipelines for deployment\n- **\ud83e\uddea Comprehensive Testing**: 93% test coverage with continuous integration\n- **\ud83d\udcca Benchmarking Tools**: Built-in performance analysis and visualization\n- **\ud83d\udd04 PyTorch Lightning**: Seamless integration with modern training workflows\n- **\ud83d\udcf1 Cross-Platform**: CPU, CUDA, MPS (Apple Silicon) support\n- **\ud83c\udf9b\ufe0f Flexible Alpha**: Fixed, learnable, or scheduled parameter control\n\n## Project Structure\n\nThe project has been organized into the following structure:\n\n```\npaGating/\n\u251c\u2500\u2500 assets/ # Static assets\n\u2502 \u2514\u2500\u2500 images/ # Image files\n\u2502 \u251c\u2500\u2500 figures/ # Paper figures\n\u2502 \u2514\u2500\u2500 plots/ # Plot outputs from experiments\n\u251c\u2500\u2500 benchmark_results/ # Results from various benchmarks\n\u2502 \u251c\u2500\u2500 coreml/ # CoreML benchmark results\n\u2502 \u251c\u2500\u2500 regression/ # Regression task results\n\u2502 \u2514\u2500\u2500 transformer/ # Transformer model results\n\u251c\u2500\u2500 coreml_models/ # Exported CoreML models\n\u251c\u2500\u2500 datamodules/ # PyTorch Lightning data modules\n\u251c\u2500\u2500 docs/ # Documentation\n\u2502 \u251c\u2500\u2500 paper/ # Research paper and references\n\u2502 \u2514\u2500\u2500 results_summary.md # Summary of experiment results\n\u251c\u2500\u2500 experiments/ # Experiment configurations\n\u251c\u2500\u2500 lightning_modules/ # PyTorch Lightning modules\n\u251c\u2500\u2500 models/ # Model implementations\n\u251c\u2500\u2500 onnx_models/ # Exported ONNX models\n\u251c\u2500\u2500 paGating/ # Core package\n\u2502 \u251c\u2500\u2500 __init__.py # Package exports\n\u2502 \u251c\u2500\u2500 base.py # Base classes\n\u2502 \u251c\u2500\u2500 paGLU.py # Gated Linear Unit implementation\n\u2502 \u251c\u2500\u2500 paGTU.py # Gated Tanh Unit implementation\n\u2502 \u251c\u2500\u2500 paSwishU.py # Swish Unit implementation\n\u2502 \u251c\u2500\u2500 paReGLU.py # ReLU Gated Linear Unit implementation\n\u2502 \u251c\u2500\u2500 paGELU.py # GELU Gated Unit implementation\n\u2502 \u251c\u2500\u2500 paMishU.py # Mish Unit implementation\n\u2502 \u251c\u2500\u2500 paSiLU.py # SiLU/Swish gating implementation\n\u2502 \u251c\u2500\u2500 paUnit.py # Generic gating unit template\n\u2502 \u2514\u2500\u2500 paGRU.py # Parameterized GRU cell\n\u251c\u2500\u2500 scripts/ # Utility scripts\n\u2502 \u251c\u2500\u2500 benchmark/ # Benchmarking scripts\n\u2502 \u2514\u2500\u2500 utilities/ # Utility scripts\n\u251c\u2500\u2500 src/ # Source code (application-specific)\n\u251c\u2500\u2500 tests/ # Test suite\n\u251c\u2500\u2500 requirements.txt # Project dependencies\n\u2514\u2500\u2500 README.md # This file\n```\n\n## Implemented Gating Units\n\n| Unit | Description | Formula |\n|------|-------------|---------|\n| paGLU | Parameterized Gated Linear Unit | x * (\u03b1 * sigmoid(x) + (1-\u03b1)) |\n| paGTU | Parameterized Gated Tanh Unit | x * (\u03b1 * tanh(x) + (1-\u03b1)) |\n| paSwishU | Parameterized Swish Unit | x * (\u03b1 * sigmoid(x) + (1-\u03b1) * x) |\n| paReGLU | Parameterized ReLU Gated Linear Unit | x * (\u03b1 * ReLU(x) + (1-\u03b1)) |\n| paGELU | Parameterized Gated GELU | x * (\u03b1 * GELU(x) + (1-\u03b1)) |\n| paMishU | Parameterized Mish Unit | x * (\u03b1 * mish(x) + (1-\u03b1)) |\n| paSiLU | Parameterized SiLU/Swish gating | x * (\u03b1 * SiLU(x) + (1-\u03b1) * x) |\n| paUnit | Generic Template for Custom Units | x * (\u03b1 * custom_fn(x) + (1-\u03b1)) |\n| PaGRUCell | Parameterized GRU Cell | Specialized recurrent architecture |\n\n## Installation\n\nClone the repository:\n```bash\ngit clone https://github.com/guglxni/paGating.git\ncd paGating\n```\n\nInstall requirements:\n```bash\npip install -r requirements.txt\n```\n\nSet up data directories and download datasets:\n```bash\npython scripts/download_data.py\n```\n\n> **Note**: This repository uses symlinks for large data files. See [docs/DATA_SETUP.md](docs/DATA_SETUP.md) for detailed setup instructions.\n\n## Quick Start\n\n### Using a paGating unit in your model\n\n```python\nimport torch\nfrom paGating import paGLU\n\n# Create a layer with fixed alpha\ngating_layer = paGLU(input_dim=512, output_dim=512, alpha=0.5)\n\n# Or with learnable alpha\nlearnable_gating_layer = paGLU(input_dim=512, output_dim=512, learnable_alpha=True)\n\n# Use in a model\nx = torch.randn(32, 512) # batch_size, input_dim\noutput = gating_layer(x) # shape: (32, 512)\n```\n\n### Integration with PyTorch models\n\n```python\nimport torch\nimport torch.nn as nn\nfrom paGating import paGLU\n\nclass MyModel(nn.Module):\n def __init__(self):\n super().__init__()\n self.fc1 = nn.Linear(784, 512)\n self.gate = paGLU(512, 512, alpha=0.5) # paGating unit\n self.fc2 = nn.Linear(512, 10)\n \n def forward(self, x):\n x = self.fc1(x)\n x = self.gate(x)\n x = self.fc2(x)\n return x\n```\n\n## Experimenting with paGating\n\n### Running Benchmarks\n\nThe framework includes tools for benchmarking different gating units:\n\n```bash\npython scripts/benchmark/benchmark_gateflow.py\n```\n\nThis generates plots comparing the performance of different units.\n\n### Running a Hyperparameter Sweep\n\nTo compare different units and alpha values:\n\n```bash\npython scripts/utilities/run_experiment_pipeline.py --experiment_name my_experiment --units paGLU paGTU paMishU --alpha_values 0.0 0.2 0.5 0.8 1.0\n```\n\nThis will:\n1. Run a hyperparameter sweep\n2. Generate a leaderboard\n3. Create visualizations\n\n### Testing with Transformer Models\n\nTo test a gating unit in a transformer for sequence classification:\n\n```bash\npython experiments/test_transformer.py --unit paMishU --alpha 0.5 --epochs 20\n```\n\n## Export to CoreML\n\nYou can export trained models to CoreML format for deployment on Apple devices:\n\n```bash\npython scripts/coreml_export.py --unit paGLU --alpha 0.5\n```\n\nTest the exported model:\n\n```bash\npython tests/test_coreml_model.py --unit paGLU --alpha 0.5\n```\n\n## Results Summary\n\nFor detailed results and comparisons of different gating units, see [docs/results_summary.md](docs/results_summary.md).\n\n## Creating Your Own Gating Unit\n\nTo create a custom gating unit:\n\n1. Create a new file in the paGating directory (e.g., `paGating/paMyCustomU.py`)\n2. Extend the `paGatingBase` class\n3. Implement the required methods\n4. Update `__init__.py` to expose your new unit\n\nExample:\n\n```python\nfrom .base import paGatingBase\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\nclass paMyCustomU(paGatingBase):\n \"\"\"\n My custom parameterized activation gating unit.\n \"\"\"\n \n def __init__(self, input_dim, output_dim, alpha=0.5, learnable_alpha=False, alpha_init=None, bias=True):\n super().__init__(\n input_dim=input_dim, \n output_dim=output_dim, \n alpha=alpha,\n learnable_alpha=learnable_alpha,\n alpha_init=alpha_init,\n bias=bias\n )\n \n def compute_gate_activation(self, x):\n # Implement your custom activation function\n return my_custom_activation(x)\n \n def forward(self, x):\n # Standard implementation, can be customized if needed\n x = self.linear(x)\n gates = self.compute_gate_activation(x)\n return x * gates\n```\n\nThen update `__init__.py`:\n\n```python\nfrom .paMyCustomU import paMyCustomU\n\n__all__ = [\n # ... existing units\n 'paMyCustomU',\n]\n```\n\n## License\n\nThis project is licensed under the Apache License 2.0 - see the [LICENSE](LICENSE) file for details.\n\n**Commercial Use**: For commercial applications, please contact the authors for licensing arrangements.\n\n## \ud83d\udcc4 Research Paper\n\nThis framework is featured in our IEEE TNNLS submission:\n\n**\"paGating: A Parameterized Activation Gating Framework for Flexible and Efficient Neural Networks for GenAI\"**\n\n- **Authors**: Aaryan Guglani, Dr. Rajashree Shettar\n- **Institution**: RV College of Engineering, Bengaluru\n- **Status**: Under Review at IEEE Transactions on Neural Networks and Learning Systems\n- **Reproducibility**: Complete reproduction guide available in [`docs/REPRODUCIBILITY.md`](docs/REPRODUCIBILITY.md)\n\n## \ud83d\udcda Documentation\n\n- **[Reproducibility Guide](docs/REPRODUCIBILITY.md)**: Step-by-step instructions to reproduce all paper results\n- **[Contributing Guide](CONTRIBUTING.md)**: How to contribute to the project\n- **[API Documentation](docs/)**: Detailed API reference and examples\n\n## \ud83c\udfc6 Citation\n\nIf you use paGating in your research, please cite:\n\n```bibtex\n@article{guglani2025pagating,\n title={paGating: A Parameterized Activation Gating Framework for Flexible and Efficient Neural Networks for GenAI},\n author={Guglani, Aaryan and Shettar, Rajashree},\n journal={IEEE Transactions on Neural Networks and Learning Systems},\n year={2025},\n note={Under Review},\n url={https://github.com/guglxni/paGating}\n}\n```\n\n## Contributing\n\nContributions are welcome! Please feel free to submit a Pull Request.\n\n## Interactive Dashboard\n\nThe project includes a Streamlit dashboard for visualizing experiment results:\n\n```bash\n# Install required packages if not already installed\npip install streamlit plotly pandas\n\n# Run the dashboard with a specific results directory\nstreamlit run scripts/streamlit_dashboard.py -- --results_dir results/your_experiment_dir\n\n# Or run the dashboard and select the results directory in the UI\nstreamlit run scripts/streamlit_dashboard.py\n```\n\nDashboard features:\n- Compare performance across different gating units\n- Analyze the effect of different alpha values\n- Explore the behavior of learnable alpha parameters\n- View training curves and leaderboards\n- Generate insights and recommendations\n\n## Experiments\n\nRun a hyperparameter sweep:\n\n```bash\npython scripts/utilities/run_experiment_pipeline.py\n```\n\nThis will:\n1. Run a sweep over different units and alpha values\n2. Generate a leaderboard\n3. Create visualizations\n4. Run the analysis\n\n## Research Paper\n\nA detailed research paper describing the paGating framework, its implementation, and experimental results is available in the [docs/paper/](docs/paper/) directory.\n",
"bugtrack_url": null,
"license": null,
"summary": "Parameterized Activation Gating Framework for Flexible and Efficient Neural Networks",
"version": "0.1.0",
"project_urls": {
"Changelog": "https://github.com/guglxni/paGating/blob/master/CHANGELOG.md",
"Documentation": "https://github.com/guglxni/paGating#readme",
"Homepage": "https://github.com/guglxni/paGating",
"Issues": "https://github.com/guglxni/paGating/issues",
"Repository": "https://github.com/guglxni/paGating"
},
"split_keywords": [
"deep-learning",
" neural-networks",
" activation-functions",
" pytorch",
" gating",
" transformer",
" machine-learning"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "fdcf681a69a796a2383765290e17219fe240df6aab65ddbb072dffc3068601af",
"md5": "930fc48656808d5e6ff2f94d940b0692",
"sha256": "5510bb64624d0c8191fe40364b0d219dc0784b09c45505f63797db86de3ba778"
},
"downloads": -1,
"filename": "pagating-0.1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "930fc48656808d5e6ff2f94d940b0692",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8",
"size": 6852316,
"upload_time": "2025-08-30T18:29:07",
"upload_time_iso_8601": "2025-08-30T18:29:07.362529Z",
"url": "https://files.pythonhosted.org/packages/fd/cf/681a69a796a2383765290e17219fe240df6aab65ddbb072dffc3068601af/pagating-0.1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "d81a6186a14081ffe47c3084b3314f2a4a9b4caef3f193cb562dd319fc2e32eb",
"md5": "ffd835c9ede32b976ecb6e28487e18ce",
"sha256": "0f0efee1040c8bc14641777852e1b6b6078bdac4da256cec6bff415b203842e7"
},
"downloads": -1,
"filename": "pagating-0.1.0.tar.gz",
"has_sig": false,
"md5_digest": "ffd835c9ede32b976ecb6e28487e18ce",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8",
"size": 6877218,
"upload_time": "2025-08-30T18:29:20",
"upload_time_iso_8601": "2025-08-30T18:29:20.812326Z",
"url": "https://files.pythonhosted.org/packages/d8/1a/6186a14081ffe47c3084b3314f2a4a9b4caef3f193cb562dd319fc2e32eb/pagating-0.1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-08-30 18:29:20",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "guglxni",
"github_project": "paGating",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"requirements": [
{
"name": "torch",
"specs": [
[
">=",
"2.0.0"
]
]
},
{
"name": "transformers",
"specs": [
[
">=",
"4.30.0"
]
]
},
{
"name": "datasets",
"specs": [
[
">=",
"2.0.0"
]
]
},
{
"name": "accelerate",
"specs": [
[
">=",
"0.20.0"
]
]
},
{
"name": "onnx",
"specs": [
[
">=",
"1.14.0"
]
]
},
{
"name": "onnxruntime",
"specs": [
[
">=",
"1.15.0"
]
]
},
{
"name": "coremltools",
"specs": [
[
">=",
"7.0"
]
]
},
{
"name": "numpy",
"specs": [
[
">=",
"1.20.0"
]
]
},
{
"name": "matplotlib",
"specs": [
[
">=",
"3.4.0"
]
]
},
{
"name": "tqdm",
"specs": [
[
">=",
"4.61.0"
]
]
},
{
"name": "pillow",
"specs": [
[
">=",
"8.2.0"
]
]
},
{
"name": "pyyaml",
"specs": []
},
{
"name": "pandas",
"specs": []
},
{
"name": "seaborn",
"specs": []
},
{
"name": "tf-keras",
"specs": [
[
">=",
"2.13.1"
]
]
}
],
"lcname": "pagating"
}
JSON
