{"id":50848403,"url":"https://github.com/larionovavi-stack/genesis2-cascade-moe","last_synced_at":"2026-06-20T07:00:38.902Z","repository":{"id":364537994,"uuid":"1268287641","full_name":"larionovavi-stack/genesis2-cascade-moe","owner":"larionovavi-stack","description":"Genesis 2 — Cascade MoE Neural Network | Patented CPU-only AI: 10,800 experts, 100% accuracy, 18ms inference, zero forgetting | No GPU required | Self-hosted alternative to GPT/LLaMA | Network automation \u0026 DevOps | Python/PyTorch","archived":false,"fork":false,"pushed_at":"2026-06-13T13:08:42.000Z","size":145,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-13T13:09:33.133Z","etag":null,"topics":["ai","artificial-intelligence","bilingual","cascade-moe","cpu-only","cybersecurity","deep-learning","devops","linux-administration","machine-learning","mixture-of-experts","network-automation","neural-network","nlp","no-gpu","on-premise","patented","pytorch","self-hosted","sysadmin"],"latest_commit_sha":null,"homepage":"https://avlarion.gumroad.com","language":"HTML","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"other","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/larionovavi-stack.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-13T11:00:22.000Z","updated_at":"2026-06-13T13:08:43.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/larionovavi-stack/genesis2-cascade-moe","commit_stats":null,"previous_names":["larionovavi-stack/genesis2-cascade-moe"],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/larionovavi-stack/genesis2-cascade-moe","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/larionovavi-stack%2Fgenesis2-cascade-moe","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/larionovavi-stack%2Fgenesis2-cascade-moe/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/larionovavi-stack%2Fgenesis2-cascade-moe/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/larionovavi-stack%2Fgenesis2-cascade-moe/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/larionovavi-stack","download_url":"https://codeload.github.com/larionovavi-stack/genesis2-cascade-moe/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/larionovavi-stack%2Fgenesis2-cascade-moe/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34560266,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-20T02:00:06.407Z","response_time":98,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai","artificial-intelligence","bilingual","cascade-moe","cpu-only","cybersecurity","deep-learning","devops","linux-administration","machine-learning","mixture-of-experts","network-automation","neural-network","nlp","no-gpu","on-premise","patented","pytorch","self-hosted","sysadmin"],"created_at":"2026-06-14T12:00:20.185Z","updated_at":"2026-06-20T07:00:38.890Z","avatar_url":"https://github.com/larionovavi-stack.png","language":"HTML","funding_links":[],"categories":["AI","Papers"],"sub_categories":["Tutorial","Further Improvement"],"readme":"\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/banner_genesis2.png\" alt=\"Genesis 2 — Cascade MoE Neural Network\" width=\"100%\"\u003e\n\u003c/p\u003e\n\n\u003ch1 align=\"center\"\u003eGenesis 2 — Cascade MoE Neural Network\u003c/h1\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003cstrong\u003eThe World's First Patented Neural Architecture That Runs on CPU\u003c/strong\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003ca href=\"#benchmarks\"\u003e\u003cimg src=\"https://img.shields.io/badge/accuracy-100%25_(111%2F111)-brightgreen?style=for-the-badge\" alt=\"Accuracy\"\u003e\u003c/a\u003e\n  \u003ca href=\"#benchmarks\"\u003e\u003cimg src=\"https://img.shields.io/badge/neurons-12,651-blue?style=for-the-badge\" alt=\"Neurons\"\u003e\u003c/a\u003e\n  \u003ca href=\"#benchmarks\"\u003e\u003cimg src=\"https://img.shields.io/badge/experts-10,800+-blue?style=for-the-badge\" alt=\"Experts\"\u003e\u003c/a\u003e\n  \u003ca href=\"#architecture\"\u003e\u003cimg src=\"https://img.shields.io/badge/GPU-not%20required-red?style=for-the-badge\" alt=\"No GPU\"\u003e\u003c/a\u003e\n  \u003ca href=\"#whats-new\"\u003e\u003cimg src=\"https://img.shields.io/badge/version-v1.1-cyan?style=for-the-badge\" alt=\"v1.1\"\u003e\u003c/a\u003e\n  \u003ca href=\"#patent\"\u003e\u003cimg src=\"https://img.shields.io/badge/patent-pending-purple?style=for-the-badge\" alt=\"Patent\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003ca href=\"https://avlarion.gumroad.com/l/lqtsbo\"\u003eAcademic $299\u003c/a\u003e \u0026bull;\n  \u003ca href=\"https://avlarion.gumroad.com/l/vrzudu\"\u003eProfessional $1,499\u003c/a\u003e \u0026bull;\n  \u003ca href=\"https://avlarion.gumroad.com/l/atmon\"\u003eEnterprise $4,999\u003c/a\u003e \u0026bull;\n  \u003ca href=\"https://avlarion.gumroad.com/l/ymyagw\"\u003eSource + Patent Bundle $5,000\u003c/a\u003e \u0026bull;\n  \u003ca href=\"https://larionovavi-stack.github.io/genesis2-cascade-moe/docs/reference-guide.html\"\u003e\u003cstrong\u003eInteractive Reference Guide\u003c/strong\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003ca href=\"https://gist.github.com/larionovavi-stack/d9bdc484813df7cb488a842cb4a0cd62\"\u003e\u003cimg src=\"https://img.shields.io/badge/Live_Demo-Try_Now-brightgreen?style=for-the-badge\" alt=\"Live Demo\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n\u003e **[Try the Live Demo](https://gist.github.com/larionovavi-stack/d9bdc484813df7cb488a842cb4a0cd62)** — click the link in the Gist for the current demo URL. Model hosted on [Kaggle](https://www.kaggle.com/datasets/alexanderlar/genesis2-cascade-moe-model). If the demo is unavailable, email **avlarionov@hotmail.com** to request a restart.\n\n---\n\n## What is Genesis 2?\n\nGenesis 2 is a **fundamentally new neural network architecture** that eliminates the need for GPU, external LLMs, and massive compute resources. It uses Cascade Activation of a Shared Neuron Pool — a patented approach where experts share neurons instead of duplicating parameters.\n\n**No GPU. No Cloud. No API costs. No token limits. Runs on your laptop.**\n\n```\nTraditional MoE:  Expert₁[500MB] + Expert₂[500MB] + ... = 50GB+, GPU required\nGenesis 2:        Expert₁[route] + Expert₂[route] + ... = 3.64 GB total, CPU only\n                  ↑ shared neuron pool, each expert is just a list of neuron IDs\n```\n\n## Why Genesis 2?\n\n| Traditional AI (GPT, LLaMA, etc.) | Genesis 2 |\n|:---|:---|\n| $2,000+/mo GPU costs | **$0** — runs on CPU |\n| API rate limits \u0026 downtime | **Unlimited** — self-hosted |\n| Data leaves your network | **100% on-premise** |\n| Catastrophic forgetting | **Zero forgetting** — mathematically guaranteed |\n| Minutes to fine-tune | **130ms** to learn a new fact |\n| Token window limits (4K-128K) | **Infinite context** — no limits |\n| Vendor lock-in | **You own the code** |\n\n## Quick Start\n\n```bash\n# Install dependencies\npip install torch numpy requests\n\n# Start the web server\npython genesis2_web.py\n\n# Open in browser\nopen http://localhost:8765\n```\n\n## API\n\n```python\nimport requests\n\nAPI = \"http://localhost:8765\"\n\n# Ask a question (returns answer + executable commands)\nr = requests.post(f\"{API}/api/query\", json={\"question\": \"configure nginx reverse proxy\"})\nprint(r.json()[\"answer\"])\nprint(r.json()[\"commands\"])\n\n# Teach new knowledge (learns in 130-550ms)\nrequests.post(f\"{API}/api/learn\", json={\n    \"question\": \"how to restart Apache\",\n    \"answer\": \"Restart Apache web server\",\n    \"exec\": \"systemctl restart apache2\"\n})\n\n# Save state\nrequests.post(f\"{API}/api/save\")\n```\n\n## What's New in v1.1 \u003ca name=\"whats-new\"\u003e\u003c/a\u003e\n\nReleased: **June 2026**\n\n| Feature | Description |\n|:--------|:------------|\n| 🧠 **Neuron Splitting** (Patent п.5) | Overloaded neurons auto-split via 2-means clustering. Coherence threshold 0.40 triggers split → two child neurons inherit parent weights |\n| 💬 **Dialogue Context** | Model tracks conversation state: \"no thanks\", \"nothing needed\", \"пока ничего\" → correct conversational replies instead of technical routing |\n| 🔧 **Command Substitution** | Auto-fills IP/port/subnet from user's question into exec commands: `ping 10.0.0.1` → `ping -c 4 10.0.0.1` |\n| 🔤 **Typo Normalization** | Repeated Cyrillic letters collapsed: \"ппривет\" → \"привет\", \"приввет\" → \"привет\" (Latin preserved: \"need\" stays \"need\") |\n| 📊 **111/111 Test Suite** | Extended benchmark from 30 to **111 queries** across 43 topics: networking, security, Docker, Cisco, VPN, DNS, databases, monitoring, SCADA, VoIP and more |\n| 🌐 **Bilingual 100%** | Both RU and EN at 100% accuracy simultaneously — verified across all 43 topic categories |\n\n## Benchmarks\n\n| Metric | v1.0 | **v1.1** |\n|:-------|:-----|:---------|\n| Shared Neurons | 12,100+ | **12,651** |\n| Trained Experts | 10,800+ | **10,800+** |\n| Test accuracy | 100% (30/30) | **100% (111/111)** |\n| Topics covered | 15 | **43** |\n| Inference latency | 18-27ms | **18-27ms** |\n| Learning speed | 130-550ms | **130ms** per fact |\n| Zero forgetting (cosine) | 1.000000 | **1.000000** |\n| Neuron splitting | ✗ | **✓ (auto)** |\n| Dialogue context | ✗ | **✓** |\n| Command substitution | ✗ | **✓** |\n| RAM usage | 3.5GB | **3.64 GB** |\n| GPU required | No | **No** |\n\n### Test Results v1.1 — 111/111 across 43 topics\n\n```\nnetworking RU/EN  ✅✅✅✅✅✅✅✅✅✅✅  (11/11)\nlinux RU/EN       ✅✅✅✅✅✅✅✅✅✅✅  (11/11)\nsecurity RU/EN    ✅✅✅✅✅✅✅✅✅  (9/9)\nvpn RU/EN         ✅✅✅✅✅  (5/5)\ndocker/k8s RU/EN  ✅✅✅✅✅✅✅✅  (8/8)\ncisco RU/EN       ✅✅✅✅✅  (5/5)\ndns/dhcp RU/EN    ✅✅✅✅✅✅  (6/6)\nmonitoring RU/EN  ✅✅✅✅✅  (5/5)\ndatabases         ✅✅✅✅  (4/4)\nnginx/web         ✅✅✅✅✅  (5/5)\nwindows           ✅✅  (2/2)\nmikrotik          ✅✅  (2/2)\nvoip/sip          ✅✅  (2/2)\nscada/iot         ✅✅  (2/2)\nbackup            ✅✅  (2/2)\ndevops            ✅✅✅✅  (4/4)\ntroubleshooting   ✅✅✅✅  (4/4)\ncloud/virt        ✅✅✅  (3/3)\nmacos             ✅✅✅  (3/3)\ntraffic           ✅✅✅  (3/3)\ngreetings/typos   ✅✅✅✅✅✅✅  (7/7)\nslang/infra       ✅✅✅✅  (4/4)\n                           ───────\nTOTAL:            ✅ 111/111 = 100%\n```\n\n## Architecture\n\nGenesis 2 is built on 8 patented innovations:\n\n### 1. Shared Neuron Pool\nAll neurons live in a single shared pool. Experts don't have their own parameters — they reference neurons by ID. One neuron can serve 50+ experts simultaneously. This makes the model **100x smaller** than traditional MoE.\n\n### 2. Expert as Route\nEach expert is just a list of neuron IDs — a \"route\" through the shared pool. Adding a new expert costs **bytes, not megabytes**. 10,800+ expert routes fit in 3.64 GB.\n\n### 3. Cascade Activation (No Router)\nTraditional MoE uses a trained router to pick experts. Genesis 2 uses a reverse index (neuron → experts) to find relevant experts in **0.14ms**. No router training, no routing errors.\n\n### 4. One-Step Learning\nTo learn a new fact: freeze all shared neurons, create a new expert with a micro-head. Takes **130-550ms**. The new knowledge never interferes with existing knowledge.\n\n### 5. Zero Catastrophic Forgetting\nEach expert has its own micro-head (output layer). New experts can't modify existing ones. **Mathematically guaranteed** — cosine similarity = 1.000000 before/after learning.\n\n### 6. Hash Neuron Embedding\nCustom embedding system with 9,761 tokens across 72 types. No dependency on external models (MiniLM, BERT, etc.). Fully self-contained.\n\n### 7. Infinite Context\nEvery learned fact becomes a permanent expert. No token window limits. 10,000 facts = 10,000 experts, all accessible instantly.\n\n### 8. Native Generation via Concept Chains\nOutput is generated through a composer that chains related concepts from activated experts. Not template matching — actual generation.\n\n```\nInput → Hash Embedding (512d) → ANN Search → Seed Experts\n     → Cascade Activation → Shared Neuron Pool → Composer → Output\n```\n\n## Knowledge Domains (35)\n\n\u003e The model is fully bilingual (RU + EN). Trained on 35 domains with 100% accuracy in both languages. Genesis 2 learns new facts in **130ms** — you can train your own model on any language and any domain in minutes, not days.\n\n\u003ctable\u003e\n\u003ctr\u003e\u003ctd\u003eNetworking (Cisco, MikroTik)\u003c/td\u003e\u003ctd\u003eLinux Administration\u003c/td\u003e\u003ctd\u003eDocker \u0026 Kubernetes\u003c/td\u003e\u003c/tr\u003e\n\u003ctr\u003e\u003ctd\u003eSecurity \u0026 Hardening\u003c/td\u003e\u003ctd\u003eWiFi Configuration\u003c/td\u003e\u003ctd\u003eDNS/DHCP/BIND\u003c/td\u003e\u003c/tr\u003e\n\u003ctr\u003e\u003ctd\u003eVPN (WireGuard, OpenVPN)\u003c/td\u003e\u003ctd\u003eDatabases (PostgreSQL, MySQL)\u003c/td\u003e\u003ctd\u003eWeb Servers (Nginx, Apache)\u003c/td\u003e\u003c/tr\u003e\n\u003ctr\u003e\u003ctd\u003eMonitoring (Zabbix, Prometheus)\u003c/td\u003e\u003ctd\u003eDevOps (Ansible, Terraform)\u003c/td\u003e\u003ctd\u003ePython Scripting\u003c/td\u003e\u003c/tr\u003e\n\u003ctr\u003e\u003ctd\u003eBash Automation\u003c/td\u003e\u003ctd\u003ePacket Analysis\u003c/td\u003e\u003ctd\u003eVoIP (Asterisk)\u003c/td\u003e\u003c/tr\u003e\n\u003ctr\u003e\u003ctd\u003eWindows Active Directory\u003c/td\u003e\u003ctd\u003emacOS Administration\u003c/td\u003e\u003ctd\u003eVirtualization\u003c/td\u003e\u003c/tr\u003e\n\u003ctr\u003e\u003ctd\u003eSCADA/ICS\u003c/td\u003e\u003ctd\u003eCloud (AWS/GCP/Azure)\u003c/td\u003e\u003ctd\u003eServer Configuration\u003c/td\u003e\u003c/tr\u003e\n\u003ctr\u003e\u003ctd\u003eMobile Protocols\u003c/td\u003e\u003ctd colspan=\"2\"\u003e\u003c/td\u003e\u003c/tr\u003e\n\u003c/table\u003e\n\n## System Requirements\n\n| Component | Minimum | Recommended |\n|:----------|:--------|:------------|\n| CPU | Any modern (ARM or x86) | 4+ cores |\n| RAM | 6 GB | 16 GB |\n| Disk | 4 GB | 10 GB |\n| Python | 3.9+ | 3.11+ |\n| PyTorch | 2.0+ | 2.3+ |\n| OS | macOS / Linux / Windows | Any |\n| GPU | **Not required** | Not required |\n\n## Patent\n\n**Status:** Filed at FIPS Russia, 31.05.2026\n**Type:** Utility Model, IPC G06N 3/04\n**Claims:** 2 independent + 6 dependent (8 total)\n**RCIS Blockchain Certificate:** #1823-376-572\n\nThe Cascade MoE architecture is protected by a pending patent. The patent covers all 8 architectural innovations listed above.\n\n## OS-Aware Execution\n\nGenesis 2 detects the host operating system and adapts:\n\n- **macOS**: Strips `sudo`, warns about Linux-only commands, uses macOS equivalents\n- **Linux**: Full command execution with `sudo` support\n- **Windows**: Suggests PowerShell alternatives\n- **Safety**: Blocks dangerous commands (`rm -rf`, `mkfs`, `dd`, `shutdown`)\n\n## Editions\n\n| Edition | Price | License | Includes |\n|:--------|:------|:--------|:---------|\n| [**Academic**](https://avlarion.gumroad.com/l/lqtsbo) | $299 | 1 person, research only | Source + model + docs |\n| [**Professional**](https://avlarion.gumroad.com/l/vrzudu) | $1,499 | 5 users, commercial | + 30 datasets + 12mo updates |\n| [**Enterprise**](https://avlarion.gumroad.com/l/atmon) | $4,999 | Unlimited, commercial | + patent docs + book + lifetime updates |\n| [**Source + Patent Bundle**](https://avlarion.gumroad.com/l/ymyagw) | $5,000 | White-label rights | + patent license + 5h consultation |\n\n## Project Structure\n\n```\ngenesis2-cascade-moe/\n├── genesis2_core.py          # Core: neurons, cascade, shared pool, training\n├── genesis2_gen.py           # Generation: concept chains, composer, boost\n├── genesis2_agent.py         # Agent: learn/reason/plan/chat/self-learn\n├── genesis2_web.py           # Web UI + REST API + OS detection\n├── genesis2_repl.py          # Interactive terminal REPL\n├── embedding/\n│   └── train_embedding.py    # Custom hash embedding training\n├── datasets/                 # 30 training datasets (Professional+)\n├── PATENT/                   # Patent materials (Enterprise+)\n└── requirements.txt\n```\n\n## Author\n\n**Larionov Alexander Viktorovich** (Ларионов Александр Викторович)\n\n- SCADA/ICS Engineer with 10+ years of industrial automation experience\n- AI Researcher specializing in novel neural architectures\n- Patent holder (Cascade MoE, FIPS Russia 2026)\n\n**Contact:** avlarionov@hotmail.com\n**GitHub:** [larionovavi-stack](https://github.com/larionovavi-stack)\n**Products:** [avlarion.gumroad.com](https://avlarion.gumroad.com)\n\n## Also by Author\n\n- **[atwSCADA](https://github.com/larionovavi-stack/awtscada)** — Free SCADA system in a single HTML file (IEC 61850, OPC UA, Modbus TCP)\n- **[Network Automation with AI](https://github.com/larionovavi-stack/network-automation-ai-guide)** — 132-page practical guide ($29)\n\n## Affiliate Program\n\nEarn **40% commission** on every sale by promoting Genesis 2.\n\n**[→ Join the Affiliate Program](https://avlarion.gumroad.com/affiliates)**\n\nPayouts via Gumroad. No approval required — instant access.\n\n## License\n\nThis repository contains the documentation, architecture description, and demo materials. The full source code and trained model are available through [Gumroad](https://avlarion.gumroad.com).\n\nPatent pending. All rights reserved. (c) 2026 Larionov Alexander Viktorovich.\n\n---\n\n\u003cp align=\"center\"\u003e\n  \u003cstrong\u003eNo GPU. No Cloud. No Limits.\u003c/strong\u003e\u003cbr\u003e\n  \u003ca href=\"https://avlarion.gumroad.com/l/lqtsbo\"\u003eGet Genesis 2 Academic — $299\u003c/a\u003e\n\u003c/p\u003e\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flarionovavi-stack%2Fgenesis2-cascade-moe","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Flarionovavi-stack%2Fgenesis2-cascade-moe","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flarionovavi-stack%2Fgenesis2-cascade-moe/lists"}