openlearning 0.0.6

A dependency installation tool for the Open-learning project

Open-learning / RGA 规则治理架构

English | 中文

---

English

Project Description

Open-learning is an open-source RGA (Rule-Governed Architecture) integration framework designed for intelligent text processing and deep learning model training. This project implements a novel rule-governed architecture that combines traditional deep learning with rule-based reasoning, providing a unique approach to neural network design.

Key Features:

• RGA Integrator: Core implementation of Rule-Governed Architecture with dynamic V-value regulation
• Smart Text Dataset: Intelligent text preprocessing with automatic vocabulary building
• Advanced Training System: Visual training progress monitoring with comprehensive metrics
• Disguise Save/Load: Save models in Transformer-compatible format for interoperability
• Memory Optimization: Automatic mixed precision and gradient checkpointing

Installation

Option 1: Install from PyPI (Recommended)

pip install openlearning

Option 2: Clone from GitHub

git clone https://github.com/Sky-zixin-yucai/Open-learning.git
cd Open-learning
pip install -e .

Quick Start

Basic Usage

from openlearning import RGAIntegrator, RGAConfig
import torch

# Create model configuration
config = RGAConfig(
    vocab_size=10000,
    dim=256,
    num_units=3  # RGA requires exactly 3 chain reaction units
)

# Initialize model
model = RGAIntegrator(config)

# Create sample input
input_ids = torch.randint(0, 10000, (1, 32))

# Forward pass
output = model(input_ids, num_cycles=3)
print(f"Logits shape: {output['logits'].shape}")
print(f"V value mean: {output['V_stats']['V_fused_mean']:.4f}")

Training Example

from openlearning import train_zixin_complete_model

# Standard training
model, history = train_zixin_complete_model(config_mode='standard')

# Quick testing mode
from openlearning import quick_test_mode
quick_test_mode()

Project Structure

openlearning/
├── __init__.py          # Module initialization and exports
├── yucai.py            # RGA integrator core implementation
├── nn.py               # Neural network components and trainers
└── pyproject.toml      # Project configuration and dependencies

Main Components

1. RGAIntegrator - Core RGA implementation with:

   • Chain reaction units with V-value regulation
   • Geological memory for multi-layer storage
   • Sandwich fusion for deep information integration
   • Formula-based validation system

2. SmartTextDataset - Intelligent dataset with:

   • Automatic character/word level detection
   • Vocabulary building with coverage statistics
   • Chinese text processing support

3. AdvancedConstrainedArchitectureTrainer - Training system with:

   • Visual progress monitoring
   • V-value health checking
   • Automatic vocabulary saving
   • Pretrained model format export

Examples

Check the example_usage() function in yucai.py for comprehensive examples including:

• Model initialization and inference
• Performance benchmarking
• Model saving and loading
• Comprehensive testing suite

Requirements

• Python >= 3.8
• PyTorch >= 1.9.0
• NumPy >= 1.19.0

Development

Install development dependencies:

pip install openlearning[dev]

License

Apache 2.0 License - See LICENSE file for details.

Contact

• Author: Open-learning Team
• Email: [e-mail]
• GitHub: https://github.com/Sky-zixin-yucai/Open-learning.git

---

中文

项目描述

Open-learning 是一个开源的 RGA（规则治理架构）集成框架，专门用于智能文本处理和深度学习模型训练。本项目实现了一种新颖的规则治理架构，将传统深度学习与基于规则的推理相结合，提供了独特的神经网络设计方法。

核心特性：

• RGA 集成器：规则治理架构核心实现，支持动态V值调控
• 智能文本数据集：自动词汇表构建的智能文本预处理
• 高级训练系统：可视化训练进度监控，包含全面指标
• 伪装保存/加载：以Transformer兼容格式保存模型，实现互操作性
• 内存优化：自动混合精度和梯度检查点技术

安装方法

方案一：通过PyPI安装（推荐）

pip install openlearning

方案二：从GitHub克隆

git clone https://github.com/Sky-zixin-yucai/Open-learning.git
cd Open-learning
pip install -e .

快速开始

基础使用

from openlearning import RGAIntegrator, RGAConfig
import torch

# 创建模型配置
config = RGAConfig(
    vocab_size=10000,
    dim=256,
    num_units=3  # RGA需要恰好3个链式反应单元
)

# 初始化模型
model = RGAIntegrator(config)

# 创建示例输入
input_ids = torch.randint(0, 10000, (1, 32))

# 前向传播
output = model(input_ids, num_cycles=3)
print(f"Logits形状: {output['logits'].shape}")
print(f"V值均值: {output['V_stats']['V_fused_mean']:.4f}")

训练示例

from openlearning import train_zixin_complete_model

# 标准训练
model, history = train_zixin_complete_model(config_mode='standard')

# 快速测试模式
from openlearning import quick_test_mode
quick_test_mode()

项目结构

openlearning/
├── __init__.py          # 模块初始化和导出
├── yucai.py            # RGA集成器核心实现
├── nn.py               # 神经网络组件和训练器
└── pyproject.toml      # 项目配置和依赖管理

主要组件

1. RGAIntegrator - RGA核心实现包含：

   • 带V值调控的链式反应单元
   • 多层级存储的地质记忆系统
   • 深度信息融合的三明治融合层
   • 基于公式的验证系统

2. SmartTextDataset - 智能数据集包含：

   • 自动字符/词级别检测
   • 带覆盖统计的词汇表构建
   • 中文文本处理支持

3. AdvancedConstrainedArchitectureTrainer - 训练系统包含：

   • 可视化进度监控
   • V值健康检查
   • 自动词汇表保存
   • 预训练模型格式导出

示例

查看 yucai.py 中的 example_usage() 函数获取全面示例，包括：

• 模型初始化和推理
• 性能基准测试
• 模型保存和加载
• 全面测试套件

系统要求

• Python >= 3.8
• PyTorch >= 1.9.0
• NumPy >= 1.19.0

开发环境

安装开发依赖：

pip install openlearning[dev]

许可证

Apache 2.0 许可证 - 详见 LICENSE 文件。

联系信息

• 作者：Open-learning 团队
• 邮箱：[e-mail]
• GitHub：https://github.com/Sky-zixin-yucai/Open-learning.git