Metadata-Version: 2.1
Name: tenIII
Version: 3.1.1
Summary: Framework avanzado de entrenamiento de modelos de Deep Learning - Ecosistema Gotham City
Author-email: Gotham City AI Lab <dev@gothamcity.ai>
Maintainer-email: Gotham City AI Lab <dev@gothamcity.ai>
License: MIT
Project-URL: Homepage, https://terminator.com.es
Project-URL: Documentation, https://terminator.com.es/docs
Project-URL: Repository, https://github.com/yoqer/tenIII
Project-URL: Issues, https://github.com/yoqer/tenIII/issues
Project-URL: Changelog, https://terminator.com.es/changelog
Keywords: deep-learning,machine-learning,training,transformer,llm,quantization,blocks,constitutional-ai,multimodal,voice,portable,usb,gotham-city,reinforcement-learning,fine-tuning,reasoning
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.8
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: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: numpy>=1.21.0
Provides-Extra: torch
Requires-Dist: torch>=2.0.0; extra == "torch"
Provides-Extra: tensorflow
Requires-Dist: tensorflow>=2.12.0; extra == "tensorflow"
Provides-Extra: blocks
Requires-Dist: safetensors>=0.3.0; extra == "blocks"
Provides-Extra: constitution
Provides-Extra: multimodal
Requires-Dist: soundfile>=0.12.0; extra == "multimodal"
Requires-Dist: librosa>=0.10.0; extra == "multimodal"
Provides-Extra: packaging
Provides-Extra: usb
Provides-Extra: quant
Requires-Dist: safetensors>=0.3.0; extra == "quant"
Requires-Dist: onnx>=1.14.0; extra == "quant"
Provides-Extra: llm
Requires-Dist: requests>=2.28.0; extra == "llm"
Requires-Dist: httpx>=0.24.0; extra == "llm"
Provides-Extra: storage
Requires-Dist: boto3>=1.26.0; extra == "storage"
Requires-Dist: requests>=2.28.0; extra == "storage"
Provides-Extra: all
Requires-Dist: torch>=2.0.0; extra == "all"
Requires-Dist: safetensors>=0.3.0; extra == "all"
Requires-Dist: soundfile>=0.12.0; extra == "all"
Requires-Dist: librosa>=0.10.0; extra == "all"
Requires-Dist: onnx>=1.14.0; extra == "all"
Requires-Dist: requests>=2.28.0; extra == "all"
Provides-Extra: dev
Requires-Dist: pytest>=7.0; extra == "dev"
Requires-Dist: pytest-cov>=4.0; extra == "dev"
Requires-Dist: black>=23.0; extra == "dev"
Requires-Dist: ruff>=0.1.0; extra == "dev"
Requires-Dist: mypy>=1.0; extra == "dev"

# TenMiNaTor III

**Framework Avanzado de Entrenamiento de Modelos de Deep Learning**

*Ecosistema Gotham City | Web: [terminator.com.es](http://terminator.com.es)*

---

## Instalación

```bash
# Instalación básica (solo Core Training + NumPy)
pip install tenIII

# Con PyTorch
pip install tenIII[torch]

# Con TensorFlow
pip install tenIII[tensorflow]

# Con Block System (dispositivos limitados)
pip install tenIII[blocks]

# Con todas las funciones
pip install tenIII[all]

# Desde web
wget http://carlomaxxine.com/terminator.com.es/downloads/teniii-3.1.1.tar.gz
pip install teniii-3.1.1.tar.gz
```

---

## Inicio Rápido

```python
from tenIII import Trainer, TrainingConfig, EarlyStoppingMode

# Configurar
config = TrainingConfig(
    model_type="transformer",
    hidden_dim=768,
    num_layers=12,
    max_epochs=100,
)

# Entrenar
trainer = Trainer(config)
model = trainer.build_model()
history = trainer.train(train_data, val_data)

# Guardar
trainer.save("model.pt")
```

---

## Características

### Core Training
- Entrenamiento estándar (SGD, Adam, AdamW, Lion)
- Fine-tuning de modelos preentrenados
- Reinforcement Learning
- Nested Learning (meta-learning)
- Entrenamiento de razonamiento
- Corrección de sesgos (steering)

### Sistema 10×12 Mejorado (Optativo)
```python
config = TrainingConfig(
    early_stopping=EarlyStoppingConfig(
        mode=EarlyStoppingMode.IMPROVED,  # Umbral relativo 0.1%
        patience=12,
        on_trigger="adapt",  # Adapta en lugar de parar
    )
)
```

Modos disponibles:
- `DISABLED`: Sin early stopping
- `CLASSIC`: Umbral absoluto (original)
- `IMPROVED`: Umbral relativo (recomendado)
- `ADAPTIVE`: Ajusta según volatilidad
- `CONTINUOUS`: Nunca para, adapta parámetros

### Control Continuo (Sin Paradas)
```python
config = TrainingConfig(
    early_stopping=EarlyStoppingConfig(
        mode=EarlyStoppingMode.CONTINUOUS,
        on_trigger="adapt",
    )
)
# El entrenamiento NUNCA se detiene
# Cuando detecta estancamiento, adapta LR, batch size, modo
```

### Hot Update (Actualización en Caliente)
```python
config = TrainingConfig(
    hot_update=HotUpdateConfig(
        enabled=True,
        update_after_10x12=True,  # Actualiza tras trigger del 10×12
    )
)

trainer = Trainer(config)
trainer.train(data)

# Desde un framework externo (estilo TenMiNaTor I):
trainer.send_hot_update({
    'type': 'weights',
    'source': 'tenminator_i_framework',
    'payload': new_weights,
})
```

### Block System (Dispositivos Limitados)
```python
config = TrainingConfig(
    blocks=BlockConfig(
        enabled=True,
        block_size=1,
        quantization=QuantizationType.Q4_0,
        freeze_blocks=[0, 1, 2],  # Congelar primeros bloques
    )
)
```

### Constitutional AI (Libre Albedrío)
```python
config = TrainingConfig(
    constitution=ConstitutionalConfig(
        enabled=True,
        policies=["honesty", "helpfulness", "harmlessness"],
        free_will_enabled=True,
        audit_enabled=True,
    )
)
```

### Multimodal (Voz)
```python
config = TrainingConfig(
    multimodal=MultimodalConfig(
        enabled=True,
        voice_enabled=True,
        thought_generator="hybrid",
    )
)
```

### USB Packaging
```python
config = TrainingConfig(
    usb_packaging=USBPackagingConfig(
        enabled=True,
        target_format="gguf",
        quantization=QuantizationType.Q4_0,
        include_anythingllm=True,
    )
)

trainer.save("./usb_output/", format="usb")
```

---

## Compatibilidad

### Backends
- **NumPy** (siempre disponible, sin dependencias)
- **PyTorch** >= 2.0
- **TensorFlow** >= 2.12

### Formatos de Modelo
- PyTorch (.pt, .bin)
- SafeTensors (.safetensors)
- GGUF (.gguf)
- TensorFlow (.h5, .pb)

### Frameworks Compatibles
- AnythingLLM
- Ollama
- llama.cpp
- LM Studio
- GPT4All
- IA-USB

---

## CLI

```bash
# Información del sistema
tenIII info

# Entrenar
tenIII train --config config.json --epochs 100

# Empaquetar para USB
tenIII package --model model.pt --output ./usb/ --quantization q4_0

# Convertir formato
tenIII convert --input model.pt --output model.gguf --format gguf

# Listar backends
tenIII backends
```

---

## Estructura del Paquete

```
tenIII/
├── core/           # Siempre incluido
│   ├── config.py   # Configuración unificada
│   ├── controller.py # Controlador continuo
│   ├── checkpoint.py # Gestión de checkpoints
│   └── trainer.py  # Trainer principal
├── backends/       # PyTorch, TensorFlow, NumPy
├── nn/             # Módulos de red neuronal
├── blocks/         # Block System (optativo)
├── constitution/   # Constitutional AI (optativo)
├── multimodal/     # Voz, KAME (optativo)
├── packaging/      # USB, GGUF (optativo)
├── storage/        # TerminaTodo (optativo)
└── cli.py          # Interfaz de línea de comandos
```

---

## Licencia

MIT License

---

## Ecosistema Gotham City

tenIII forma parte del ecosistema Gotham City de herramientas de IA:

- **tenIII** - Framework de entrenamiento (este paquete)
- **TenMiNaTor** - Versiones anteriores (I, II)
- **TerminaTodo** - Gestión de almacenamiento
- **TERMINATORI** - Framework de interacción

---

*Desarrollado por Gotham City AI Lab*
*Web: [terminator.com.es](http://terminator.com.es)*
