Metadata-Version: 2.4
Name: universal-model
Version: 0.1.2
Summary: A lightweight, CPU-first universal neural model framework built on CPUTorch
Author: Frandika Imam Arifin
License: MIT
Project-URL: Homepage, https://github.com/Mangono1/universal_model
Project-URL: Repository, https://github.com/Mangono1/universal_model
Project-URL: Issues, https://github.com/Mangono1/universal_model/issues
Keywords: neural-network,deep-learning,machine-learning,cpu,cputorch,model-framework
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Operating System :: OS Independent
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Description-Content-Type: text/markdown
Requires-Dist: cputorch>=0.5.5

# Universal Model Framework

**Universal Model Framework** is a modular neural-model framework built on top of **CPUTorch**.

The framework is designed to provide the model layer above the computational engine.

## Architecture

```text
Universal Model Framework
          |
          v
       CPUTorch
          |
          v
    Tensor / Compute
CPUTorch is the required computational foundation.
The framework does not require a predefined domain dataset.
Users can create their own models and datasets for:
Agriculture
Medicine
Science
Education
Engineering
Finance
Language
Custom domains
Design Principles
1. CPUTorch is the foundation
Universal Model Framework depends on CPUTorch.
The CPUTorch source code is not copied into this project.
2. Domain independent
The framework does not force a specific dataset or knowledge domain.
3. Modular
Different components can evolve independently:
Model
Layers
Dataset
Tokenizer
Trainer
Optimizer
Loss
Evaluation
Checkpoint
4. Scalable
The architecture is designed to support small models as well as larger models, including models targeting approximately 100 million parameters.
Installation
python -m pip install universal-model
CPUTorch will be installed as a dependency.
Development
python examples/basic_model.py
Project Structure
universal_model/
├── core/
├── layers/
├── models/
├── data/
├── training/
├── losses/
├── evaluation/
└── utils/
Roadmap
V0.1
CPUTorch integration foundation
Module abstraction
Parameter abstraction
Model configuration
Base model
Project packaging
V0.2
Linear layer
Embedding
Activations
Normalization
V0.3
Attention
Transformer blocks
Transformer model
V0.4
Dataset API
DataLoader
Tokenizer interface
V0.5
Trainer
Optimizers
Scheduler
Checkpointing
V0.6
Evaluation
Metrics
Parameter counting
Model inspection
V0.7
Target-parameter architecture generation
V1.0
Universal Model Framework.
Author
Frandika Imam Arifin
License
MIT EOF

**Perhatikan:** `EOF` terakhir harus sendirian di barisnya:

```text
