Skip to content

REPRODUCE.md

Latest commit

 

History

History
270 lines (220 loc) · 7.31 KB

REPRODUCE.md

File metadata and controls

270 lines (220 loc) · 7.31 KB
Raw

Reproducing Paper Results

This document provides step-by-step instructions to reproduce all results reported in the paper:

"The Dialectical Protocol: Self-Evolving AI via Bio-Inspired Vibe Coding and Adversarial Agent Synthesis"

Estimated Time: 6-8 hours (mostly unattended)

🖥️ Hardware Requirements

Minimum Configuration

  • CPU: Intel Core Ultra 7 255H (or equivalent 16-core)
  • GPU: Intel Arc 140T (8 Xe-cores, 74 TOPS)
  • NPU: Intel AI Boost NPU 3 (13 TOPS)
  • RAM: 32 GB DDR5
  • Storage: 100 GB free

Tested Configurations

  • ✅ Samsung Galaxy Book5 Pro (NP965XHD)
  • ✅ ASUS Zenbook with Intel Arc iGPU
  • ⚠️ NVIDIA GPUs: Compatible but power monitoring unavailable

💻 Software Requirements

Component Version Notes
OS Windows 11 25H2 / Ubuntu 22.04
Python 3.11.8 Exact version recommended
Ollama 0.1.x Local LLM inference
OpenVINO 2026.0.0 NPU support
Intel oneAPI 2025.0 (Optional) GPU optimization

📦 Installation (15 minutes)

Step 1: Clone Repository

git clone https://github.com/yourusername/BioNexus-Dialectical-Protocol.git
cd BioNexus-Dialectical-Protocol

Step 2: Environment Setup

For Intel Arc (Windows):

# Set environment variables
[System.Environment]::SetEnvironmentVariable("ZES_ENABLE_SYSMAN","1","User")
[System.Environment]::SetEnvironmentVariable("OLLAMA_INTEL_GPU","true","User")

# Install Python dependencies
pip install -r requirements.txt

For Intel Arc (Linux):

export ZES_ENABLE_SYSMAN=1
export OLLAMA_INTEL_GPU=true
pip install -r requirements.txt

Step 3: Download LLM Models (10 minutes)

ollama pull llama3.2:3b-instruct-q4_K_M
ollama pull mistral-large-2

Step 4: Verify Installation

python -c "from core import dialectical_protocol; print('✓ Import successful')"
python benchmarks/verify_environment.py

Expected Output:

✓ Python 3.11.8
✓ OpenVINO 2026.0.0 detected
✓ NPU available (Intel AI Boost NPU 3)
✓ Ollama running
✓ Model llama3.2:3b loaded

🧪 Reproducing Main Results

Result 1: Evolution Trajectory (Table 1, Figure 2)

Command:

python main.py --mode evolve --generations 50 --trials 5 --output results/evolution/

Expected Runtime: ~4 hours

Expected Output (results/evolution/summary.json):

{
  "generation_0": {"accuracy": 0.620, "std": 0.015},
  "generation_5": {"accuracy": 0.683, "std": 0.012},
  "generation_10": {"accuracy": 0.741, "std": 0.009},
  "generation_14": {"accuracy": 0.872, "std": 0.007},
  "generation_30": {"accuracy": 0.913, "std": 0.005},
  "generation_50": {"accuracy": 0.928, "std": 0.004}
}

Validation:

  • Gen 0 accuracy should be 62.0% ± 1.5%
  • Gen 14 should show Bayesian discovery (spike to 87.2%)
  • Gen 50 should reach 92.8% ± 0.5%

Result 2: Vibe Coding vs Traditional Prompting (Table 2, Figure 3)

Command:

python benchmarks/ablation_study.py --methods vibe,traditional,hybrid --trials 5

Expected Runtime: ~2 hours

Expected Output:

Method                  Success Rate    Compile Failures
Traditional Prompting   68% ± 3%        24% ± 2%
Vibe Coding            94% ± 2%         4% ± 1%
Hybrid (AIBRA Peak)    98% ± 1%         0%

Result 3: SAC Hallucination Detection (Section 4.1, Figure 4)

Protocol: Uses 850 logical inconsistencies (30% Syntactic, 70% Semantic context mashups). Command:

python benchmarks/sac_validation.py --dataset benchmarks/multi_omics_conflict/

Expected Runtime: ~30 minutes

Expected Output:

SAC Performance (127 test cases):
- True Positive:  120 (94.5%)
- False Positive:   3 (2.4%)
- False Negative:   4 (3.1%)
- Mean L_C (Hallucinations): 0.847 ± 0.12
- Mean L_C (Valid Reports):  0.32 ± 0.18

Result 4: Case Study - Pathogen Detection in Gen 23 (§5.1)

Command:

python benchmarks/ajtert_case_study/reproduce_gen23.py

Expected Runtime: ~5 minutes

Expected Output:

═══════════════════════════════════════════
SAC Validation Report
═══════════════════════════════════════════
Valid: False
Consensus Loss (L_C): 0.847
  - Source Divergence: 0.950
  - Context Entropy: 0.682

Flagged Entity Pairs: 1
  • AjTERT ↔ Fibonacci_Seeding

[!WARNING] Hallucination Detected
─────────────────────────────────
This report merges contexts from 2 projects:
  - SeaCucumber_Genomics
  - Wagyu_Scaling

Entity pairs with ZERO co-occurrence:
  • AjTERT ↔ Fibonacci_Seeding
═══════════════════════════════════════════

Result 5: Digital Thymus (Appendix E, Figure 5)

Command:

python core/digital_thymus/demo.py --register-gen23

Expected Output:

Digital Thymus — Immune System Status
═════════════════════════════════════
Total Hallucinations:  1
Success Rate:          100%
Knowledge Integrity:   98.2%
Experience Tokens:     1.0
═════════════════════════════════════

Recent Incidents:
  [2026-03-09-001] AjTERT ↔ Fibonacci_Seeding
    Gen 23 | L_C=0.847 | Status: MITIGATED

📊 Generating Figures

Command:

python scripts/generate_figures.py \
  --input results/ \
  --output paper/figures/ \
  --format pdf

Generated Files:

  • paper/figures/fig2_evolution.pdf
  • paper/figures/fig3_comparison.pdf
  • paper/figures/fig4_sac.pdf
  • paper/figures/fig5_thymus.pdf

🐛 Troubleshooting

  • Issue: "NVML library not found"

    • Cause: Intel Arc does not support NVIDIA NVML
    • Solution: Use --no-power flag
    python main.py --mode evolve --no-power

  • Issue: "Generation takes > 10 minutes"

    • Cause: Ollama not using GPU acceleration
    • Solution: Check environment variables
    echo $OLLAMA_INTEL_GPU  # Should be "true"
ollama ps  # Check "processor: gpu"

  • Issue: "Accuracy differs by > 2%"

    • Cause: Stochastic LLM sampling
    • Solution: This is expected. Run with --seed 42 for deterministic results
    python main.py --mode evolve --seed 42

  • Issue: "OpenVINO NPU not detected"

    • Cause: NPU driver not installed
    • Solution:
      • Windows: Install Intel NPU driver from Intel website
      • Linux: Install intel-level-zero-gpu package

✅ Validation Checklist

  • Gen 0 accuracy: 62.0% ± 1.5%
  • Gen 14 Bayesian discovery observed
  • Gen 50 accuracy: 92.8% ± 0.5%
  • Vibe Coding: 94% ± 2% success
  • SAC detection: 94% ± 2%
  • Gen 23 case: L_C = 0.847
  • Digital Thymus: 100% self-correction
  • All figures generated successfully

📧 Support

If you encounter issues:

  1. Check FAQ
  2. Open GitHub Issue
  3. Email: support@bionexus-lab.org

Last Updated: 2026-03-09 Tested Platforms: Windows 11, Ubuntu 22.04 Maintainer: [Your Name]