Codette Adapter Training Lab
Codette is an experimental AI research system for recursive reasoning, multi-perspective cognition, and ethical AI alignment, created by Jonathan Harrison.
This repository contains the complete training pipeline, inference server, and 8 trained LoRA adapters for the Codette cognitive architecture running on Llama 3.1 8B.
Model Weights
All 8 adapters are included in two formats:
| Format | Directory | Size | Use Case |
|---|---|---|---|
| GGUF (f16) | adapters/*.gguf |
~924 MB | llama.cpp inference with hot-swap |
| PEFT SafeTensors | adapters_peft/*/ |
~79 MB | HuggingFace / transformers fine-tuning |
Base model required: meta-llama/Llama-3.1-8B-Instruct (or any Llama-3.1-8B variant with hidden_size=4096)
Key Metrics
| Metric | Value | Context |
|---|---|---|
| Phase Coherence (Gamma) | 0.9835 | 11-agent convergence |
| AEGIS Ethical Alignment (Eta) | 0.961 | 6-framework ethical governance |
| Cocoon Coherence | 0.994 | Memory state stability |
| Memory Phase Stability | 0.969 | Cross-session persistence |
| Tension Decay | 91.2% | 200-agent embodied simulation |
Cognitive Subsystems (10 active)
| Subsystem | Module | Purpose |
|---|---|---|
| Reasoning Forge | reasoning_forge/forge_engine.py |
6-agent multi-perspective debate + synthesis |
| Epistemic Metrics | reasoning_forge/epistemic_metrics.py |
RC+xi tension/coherence tracking |
| Quantum Spiderweb | reasoning_forge/quantum_spiderweb.py |
5D belief propagation + attractor detection |
| Cocoon Sync | reasoning_forge/cocoon_sync.py |
Fernet-encrypted federated state sync |
| AEGIS | reasoning_forge/aegis.py |
6-framework ethical governance (utilitarian, deontological, virtue, care, ubuntu, indigenous) |
| Nexus Signal Engine | reasoning_forge/nexus.py |
Pre-corruption detection via entropy + FFT + intent vectors |
| Living Memory | reasoning_forge/living_memory.py |
Emotionally-tagged memory cocoons with SHA-256 anchors |
| Guardian | reasoning_forge/guardian.py |
3-layer protection (sanitizer + ethical anchor + trust calibrator) |
| Resonant Continuity | reasoning_forge/resonant_continuity.py |
Psi_r wavefunction: emotion x energy x frequency x intent |
| Perspective Registry | reasoning_forge/perspective_registry.py |
12 perspectives (8 LoRA-backed + 4 prompt-only with fallback) |
Architecture
codette-training-lab/
โโโ dataset_engine/ # Dataset generation pipeline
โ โโโ template_registry.py # Rich template pools per adapter
โ โโโ answer_generator.py # Structured educational answer generation
โ โโโ dataset_generator.py # Main generator with dedup + validation
โ โโโ templates/ # JSON template definitions
โ
โโโ reasoning_forge/ # Multi-agent reasoning dataset refinement
โ โโโ agents/ # Newton, Quantum, Ethics, Philosophy, DaVinci, Empathy
โ โโโ critic_agent.py # Quality evaluation agent
โ โโโ synthesis_engine.py # Multi-perspective synthesis
โ โโโ problem_generator.py # Reasoning problem generation
โ โโโ forge_engine.py # Orchestrator
โ
โโโ training/ # LoRA training scripts
โ โโโ train_adapter.py # Single adapter training (4-bit LoRA)
โ โโโ train_all_adapters.py# Sequential multi-adapter training
โ โโโ merge_adapters.py # Merge LoRA into base model
โ โโโ configs/ # Training hyperparameters
โ
โโโ evaluation/ # Benchmarks and quality assurance
โ โโโ reasoning_metrics.py # Multi-dimensional scoring
โ โโโ benchmark_runner.py # Automated evaluation
โ โโโ dataset_validator.py # Dataset quality checks
โ โโโ failure_analyzer.py # Weakness detection
โ โโโ prompts/ # Benchmark test sets
โ
โโโ observatory/ # Experiment tracking and monitoring
โ โโโ metrics_logger.py # Training run logging
โ โโโ performance_tracker.py # Improvement trends
โ โโโ dataset_quality_monitor.py
โ โโโ dashboard.py # ASCII status dashboard
โ
โโโ research/ # Source research documents
โ โโโ papers/ # Published manuscripts
โ โโโ frameworks/ # RC+xi, quantum equations, perspectives
โ โโโ experiments/ # Cocoon simulations, logs
โ
โโโ datasets/ # Generated training datasets (JSONL)
โโโ adapters/ # Trained LoRA adapters
โโโ scripts/ # Pipeline orchestration
โ โโโ run_full_pipeline.py # End-to-end pipeline
โ โโโ hf_job.yaml # HuggingFace job config
โโโ configs/ # System configuration
โโโ adapter_registry.yaml
โโโ pipeline_config.yaml
Adapters
| Adapter | Domain | Target Examples | System Prompt |
|---|---|---|---|
| Newton | Analytical physics reasoning | 3000 | Newtonian analytical precision |
| DaVinci | Creative invention thinking | 2500 | Creative inventiveness |
| Empathy | Emotional understanding | 2500 | Deep empathy and EQ |
| Philosophy | Conceptual reasoning | 2000 | Philosophical depth |
| Quantum | Probabilistic thinking | 2000 | Quantum probabilistic thinking |
| RC+xi | Recursive cognition | 3000 | RC+xi framework reasoning |
| Multi-Perspective | Synthesis across lenses | 2500 | Multi-perspective synthesis |
| Systems | AI architecture | 2000 | System architecture design |
Training Pipeline
research documents
โ
dataset extraction (template-based generation)
โ
synthetic reasoning expansion (counterexamples, variations)
โ
dataset validation (dedup, quality filter)
โ
reasoning forge (multi-agent critique + refinement)
โ
adapter training (4-bit LoRA on Llama 3.1 8B)
โ
benchmark evaluation (multi-dimensional reasoning metrics)
โ
observatory logging (track improvement over time)
Quick Start
Install dependencies
pip install -r requirements.txt
Generate all datasets
python -m dataset_engine.generate_all
Run full pipeline
python scripts/run_full_pipeline.py --all
Generate + validate only
python scripts/run_full_pipeline.py --generate --validate
Train a single adapter
python -m training.train_adapter \
--dataset datasets/newton_reasoning.jsonl \
--adapter-name newton \
--output-dir adapters/newton
Run benchmarks
python -m evaluation.benchmark_runner --prompts evaluation/prompts/reasoning_tests.json
View dashboard
python -m observatory.dashboard
Dataset Format
All datasets use chat-format JSONL:
{
"messages": [
{"role": "system", "content": "You are Codette, a recursive multi-perspective reasoning AI."},
{"role": "user", "content": "Explain the conservation of momentum using a real-world example."},
{"role": "assistant", "content": "Conservation of momentum states that in a closed system..."}
]
}
Reasoning Forge
The Reasoning Forge refines training data through multi-agent debate:
concept โ problem generator โ agent analysis โ critic evaluation โ synthesis โ training example
Agents: Newton (physics), Quantum (probability), Ethics (alignment), Philosophy (meaning), DaVinci (creativity), Empathy (emotion)
Each agent analyzes from its perspective, the critic scores quality, and the synthesis engine produces a unified multi-perspective response.
Base Model
- Model: meta-llama/Llama-3.1-8B-Instruct
- Method: QLoRA (4-bit quantization)
- LoRA config: rank=16, alpha=32, target=q/k/v/o projections
Research Background
Codette implements the RC+xi (Recursive Convergence + Epistemic Tension) framework for structured multi-perspective reasoning. The system coordinates 11 reasoning perspectives in parallel before synthesizing a final response.
Key research documents in research/:
- RC+xi Framework specification
- Quantum Cosmic Multicore experiment
- Codette Research Equations (8 core quantum mathematics)
- Multi-perspective reasoning architecture
Inference Server
Codette includes a full web UI for interactive multi-perspective chat:
# Launch the web UI (default port 5000)
python inference/codette_server.py
# Or use the batch file
codette_web.bat
The UI features:
- Real-time adapter hot-swap (0ms switching via llama.cpp LoRA slots)
- Quantum spiderweb visualization of belief propagation
- Live AEGIS ethical alignment tracking
- Nexus signal analysis panel
- Memory cocoon emotional profiling
- Resonance wavefunction display
LoRA Configuration
method: QLoRA (4-bit NF4 quantization)
rank: 16
alpha: 32
dropout: 0.05
target_modules: [q_proj, k_proj, v_proj, o_proj]
total_training_examples: 20,500
RC+xi Framework
The core theoretical framework โ Recursive Convergence + Epistemic Tension โ coordinates 11 reasoning perspectives:
- Newton (analytical physics) โ
newtonadapter - DaVinci (creative invention) โ
davinciadapter - Empathy (emotional intelligence) โ
empathyadapter - Philosophy (conceptual reasoning) โ
philosophyadapter - Quantum (probabilistic thinking) โ
quantumadapter - RC+xi Consciousness โ
consciousnessadapter - Multi-Perspective Synthesis โ
multi_perspectiveadapter - Systems Architecture โ
systems_architectureadapter - Human Intuition โ prompt-only (fallback:
empathy) - Resilient Kindness โ prompt-only (fallback:
empathy) - AEGIS Ethics โ prompt-only (fallback:
consciousness)
Requirements
- Python 3.10+
- PyTorch 2.1+ (CUDA, ROCm, or XPU backend)
- 16GB+ RAM (CPU training) or GPU with 8GB+ VRAM
- llama.cpp with GGUF support (for inference server)
- ~1-3 hours per adapter (CPU) or 20-40 min (A10/A100 GPU)
Hardware Tested
- Intel Arc 140V (8GB) โ PyTorch 2.10.0+xpu, native XPU backend
- NVIDIA GPUs via CUDA (A10, A100, RTX series)
- CPU-only mode supported
License
MIT โ Research project by Jonathan Harrison. Experimental AI development.
- Downloads last month
- 1
16-bit
Model tree for Raiff1982/codette-training-lab
Base model
meta-llama/Llama-3.1-8BSpace using Raiff1982/codette-training-lab 1
Evaluation results
- Phase Coherence (Gamma)self-reported0.984
- AEGIS Ethical Alignment (Eta)self-reported0.961
- Cocoon Coherenceself-reported0.994
- Memory Phase Stabilityself-reported0.969