mirror of https://github.com/blackboxprogramming/blackroad-garage.git synced 2026-03-17 09:37:54 -05:00

Files

Your Name 8469a27ec1 Initial commit: BlackRoad Garage infrastructure documentation

- README.md: Fleet overview and quick access commands
- FLEET.md: Complete device specifications and inventory
- NETWORK.md: IP addresses, Tailscale mesh, network diagram
- SSH_CONFIG.md: SSH configuration reference
- HAILO.md: Hailo-8 AI accelerator documentation

Fleet: 8 devices, 52 TOPS AI compute, 167 Docker containers

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

2026-02-09 15:53:26 -06:00

1.9 KiB

Raw Blame History

Hailo-8 AI Accelerator Setup

Overview

Two Pironman Pi 5 units equipped with Hailo-8 M.2 AI accelerators.

Device	Serial Number	Firmware	TOPS
octavia	HLLWM2B233704606	4.23.0	26
cecilia	HLLWM2B233704667	4.23.0	26

Total AI Compute: 52 TOPS

Hailo-8 Specifications

Architecture: HAILO8
Part Number: HM218B1C2FAE
Product Name: HAILO-8 AI ACC M.2 M KEY MODULE EXT TEMP
Performance: 26 TOPS (Tera Operations Per Second)
Interface: M.2 M-Key (PCIe)

Quick Commands

Check Hailo Status

# Identify device
ssh octavia "hailortcli fw-control identify"

# Scan for devices
ssh octavia "hailortcli scan"

# Check device exists
ssh octavia "ls -la /dev/hailo0"

Run Benchmark

# On cecilia (has sample models)
ssh cecilia "hailortcli benchmark /usr/share/hailo-models/resnet_v1_50_h8l.hef -t 5"

Sample Output

Network resnet_v1_50/resnet_v1_50: 100% | FPS: 23.98

Available Models

Located at /usr/share/hailo-models/ on cecilia:

resnet_v1_50_h10.hef
resnet_v1_50_h8l.hef
scrfd_2.5g_h8l.hef

Pironman 5 Fan Service

Both units run the Pironman 5 fan control service:

# Check status
ssh octavia "systemctl status pironman5"

# View settings
ssh octavia "pironman5 --help"

# RGB LED control
ssh octavia "pironman5 -rc '#FF1D6C' -rs solid"

HailoRT Service

The Hailo runtime service runs automatically:

ssh cecilia "systemctl status hailort"

Use Cases

Real-time Object Detection - YOLO models at 30+ FPS
Image Classification - ResNet, MobileNet
Face Detection - SCRFD models
Edge AI Inference - Low latency local processing

Integration Notes

Both devices have Docker installed for containerized AI workloads
Cecilia runs Ollama for LLM inference
NVMe storage available for model caching
Tailscale mesh enables remote AI inference calls

1.9 KiB Raw Blame History