COGNITIVE-CORES Framework

🧠 Universal Standard for Cognitive Architectures by Ame Web Studio

Cognitive-Cores is a robust, agnostic framework designed for building advanced cognitive AI models. It provides a standardized interface for integrating Vision, Language, Audio, World Modeling, and Multimodal capabilities into a unified system.

🚀 Installation

Option 1: Via Pip (From PyPI)

pip install cognitive-cores

Option 2: Via Pip (From GitHub)

pip install git+https://github.com/Volgat/nexus-standardisation.git@cognitive-core

Option 3: Via HuggingFace

pip install git+https://huggingface.co/amewebstudio/cognitive-core

Optional Dependencies

pip install "cognitive-cores[vision]"    # For Vision Models
pip install "cognitive-cores[audio]"     # For Audio Models
pip install "cognitive-cores[training]"  # For Training Tools (WandB, etc.)
pip install "cognitive-cores[all]"       # Full Installation

🛠️ Usage

Loading Models regarding Cognitive Finetuning

To finetune a model built with Cognitive-Cores (like NEXUS-LPOL) from HuggingFace, use the standard AutoModel interface with trust_remote_code=True.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from cognitive_core import CognitiveTrainer, CognitiveTrainingConfig, prepare_dataset

# 1. Configuration
model_id = "amewebstudio/nexus-lpol-v3"  # Example Model

# 2. Load Tokenizer & Model
# trust_remote_code=True is essential to load the custom cognitive architecture
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True, 
    torch_dtype=torch.float16,
    device_map="auto"
)

# 3. Training Setup
config = CognitiveTrainingConfig(
    output_dir="./nexus-finetuned",
    num_train_epochs=3,
    per_device_train_batch_size=4
)

# 4. Initialize Trainer
trainer = CognitiveTrainer(
    model=model,
    args=config,
    train_dataset=my_dataset, # Prepare your dataset using prepare_dataset helper
)

# 5. Start Finetuning
trainer.train()

🧩 Core Capabilities

The framework provides a suite of standardized, reusable modules designed for high-performance cognitive modeling.

• Advanced Normalization & Encoding: Optimized implementations for stability and long-context handling.
• Attention Mechanisms: Efficient attention layers supporting extensive context windows and multimodal fusion.
• Memory Systems: sophisticated short-term, long-term, and episodic memory modules.
• World Modeling: Components for simulating and predicting states across physical, social, and abstract domains.
• Internal State Management: Modules for handling agentic internal states, drives, and cohesion.
• Multimodal Integration: Universal latent space mapping for seamless alignment of text, audio, and visual data.
• Neurogenesis: Dynamic architectural adaptation capabilities.

📄 License

PROPRIETARY - ALL RIGHTS RESERVED Copyright © 2026 Mike Amega - Ame Web Studio