Consumer GPU Node Setup and Steam Tool

🎮 Consumer GPU Node Setup & Free Steam RTX Tool (v1.8.0)

This guide walks through setting up a consumer gaming PC (6 GB – 24 GB VRAM, or CPU fallback) as a HiveMind training miner, storage seeder, and local OpenAI-compatible inference gateway, and running the opt-in Community GPU Benchmark Suite.

⚖️ Non-Monetary Charter (0.00 EUR / 0.00 USD): Running a node or contributing code involves zero financial payment. Accumulated "Points" (PTS / $HIVE) confer solely computing processing benefits (priority Sparse MoE inference tokens, speculative decoding bandwidth, and distributed QLoRA fine-tuning slots) when the live production HiveMind Swarm is deployed, available, and used by a large community.

---

1. Installation & Environment Setup

# Clone Project HiveMind and create an isolated virtual environment
git clone https://github.com/xaviercallens/HiveMind.git
cd HiveMind
uv venv
uv pip install -e ".[dev]"

---

2. Opt-In Community GPU Benchmark (< 1.5 GB Peak VRAM)

Test your GPU (NVIDIA RTX 20/30/40/50, AMD Radeon RX 6000/7000 ROCm, Apple Silicon M1–M4 MPS, or CPU fallback) on real-token QLoRA convergence, EF21-DiLoCo residual contraction, 128-expert 3-tier MoE routing, and ZK-OPoI v4 batch Merkle commitments:

# Run the 5-stage benchmark and generate artifacts/community_gpu_benchmark_receipt.json
uv run python scripts/run_community_gpu_benchmark.py --contributor-handle <your-github-handle>

# Run the automated pytest suite
uv run pytest tests/test_community_gpu_benchmark.py -v

Share your signed receipt_sha256 in Discussion #84 (Community GPU Benchmark & Hardware Receipts Board)!

---

3. CLI Commands & Local Servers

Command	Description	Default Endpoint / Output
python scripts/run_community_gpu_benchmark.py --contributor-handle <user>	Runs the 5-stage opt-in Community GPU Benchmark (< 1.5 GB VRAM) and emits a SHA-256 signed receipt	artifacts/community_gpu_benchmark_receipt.json
python -m hivemind.cli.main hardware-check	Detects GPU vendor (cuda, rocm, mps, cpu), free VRAM budget, and Win32 Idle Daemon state	CLI Output
python -m hivemind.cli.main deploy-steam-tool	Runs the 10-Stage Intense RTX GPU Stress Suite and packages steam_dist/HiveMind_Steam_FreeEdition/	artifacts/
python -m hivemind.cli.main play-hack-game --season s1-s6	Executes the Hugging Face Hack-Miner Campaign across Missions 1–24 (SelfPlayCodeRepairArena & Lean4AutoformalizerArena)	artifacts/hf_hack_miner_game_report.json
python -m hivemind.cli.main serve-hack-hud --port 8090	Launches the Cyberpunk RTX Control Center, Hollow uRing WebGL 3D Geo-Ring Globe (PR #70), Season 6 Agent Arena & Strategy HUD (PR #78)	http://127.0.0.1:8090
python -m hivemind.cli.main serve-coordinator --port 8000	Starts the BOINC Coordinator (FL-08 hardened, PR #72) & OpenAI-compatible /v1/chat/completions (stream=True SSE) gateway	http://127.0.0.1:8000/v1

---

4. Connecting Cursor, Continue.dev & OpenAI Clients

Point any OpenAI-compatible IDE plugin or SDK to http://127.0.0.1:8000/v1:

{
  "models": [
    {
      "title": "HiveMind Sparse MoE Swarm (Local P2P Gateway)",
      "provider": "openai",
      "model": "hivemind-qwen2.5-sparse-moe-top2",
      "apiBase": "http://127.0.0.1:8000/v1",
      "apiKey": "hivemind-local-pouw-credit"
    }
  ]
}