Colossus —
Multi-System Temporal Convolutional Coordinator
Colossus coordinates all mechanical systems in large commercial facilities — boilers, chillers, air handlers, and distribution pumps — into a unified optimization target. Named for the ancient wonder of Rhodes. The largest and most complex TCN in the fleet, with the highest parameter count and deepest effective receptive field.
Abstract
Background
Colossus coordinates all mechanical systems in large commercial facilities — boilers, chillers, air handlers, and distribution pumps — into a unified optimization target. Named for the ancient wonder of Rhodes. The largest and most complex TCN in the fleet, with the highest parameter count and deepest effective receptive field.
Approach
The model was trained in AxonML (a pure-Rust deep learning framework) and compiled through the Hailo Dataflow Compiler (DFC 3.33.1) targeting Hailo-8 silicon. Post-training INT8 quantization was applied during the DFC compilation pass with production telemetry calibration data. The resulting Hailo Executable Format (HEF) binary executes on Hailo’s fixed-function dataflow architecture with deterministic latency and zero framework overhead at the edge.
Results
On production hardware (Hailo-8 M.2 (P/N: HM218B1C2FAE, S/N: HLDDM2A234600289)), Colossus achieves 457 FPS (hw_only) with 2.209 ms hardware latency at 2.22 W average power draw.
Conclusion
Colossus is production-ready as a single HEF binary deployed to edge devices with no external dependencies beyond the HailoRT vendor runtime. The model meets real-time latency requirements for its target hvac control application.
| Model | Colossus |
| Domain | HVAC Control |
| Architecture | Multi-System Temporal Convolutional Coordinator |
| Target silicon | Hailo-8 |
| Measured on | Hailo-8 M.2 (P/N: HM218B1C2FAE, S/N: HLDDM2A234600289) |
| DFC compiler | 3.33.1 |
| Framework | AxonML v0.6 (pure-Rust, CUDA + CPU backends) |
| Author | Andrew Jewell Sr. · ORCID 0009-0005-2158-7060 |
| Organization | AutomataNexus LLC · Fort Wayne, Indiana |
Executive overview
Colossus coordinates all mechanical systems in large commercial facilities — boilers, chillers, air handlers, and distribution pumps — into a unified optimization target. Named for the ancient wonder of Rhodes. The largest and most complex TCN in the fleet, with the highest parameter count and deepest effective receptive field.
Network I/O
Input: 48-channel facility-wide telemetry aggregate. Output: system-level optimization targets (CW setpoint, HW setpoint, SAT setpoints, staging commands).
Architecture
Multi-System Temporal Convolutional Coordinator
8-layer deep dilated TCN with the widest input tensor in the fleet (48 channels). Ingests aggregated telemetry from all building subsystems simultaneously: chiller plant status, boiler plant status, AHU supply temps, zone temperature averages, outdoor conditions, electrical demand, gas consumption, and scheduling inputs. The model predicts whole-building energy optimization targets that downstream controllers (Aquilo, Boreas, Vulcan, etc.) use as setpoint constraints. Implements hierarchical predictive control where Colossus sets the strategy and site-specific models execute tactically.
Compilation constraints
All AxonML models targeting Hailo silicon are compiled under the fixed-function dataflow constraints: no dynamic control flow, no variable-length dimensions, all activations representable in INT8 after calibration, and no operations requiring dedicated softmax hardware (replaced with ReLU gating or depthwise convolution equivalents where necessary).
Silicon performance
Measured on production hardware via hailortcli benchmark with 5-second sustained inference. Device: Hailo-8 M.2 (P/N: HM218B1C2FAE, S/N: HLDDM2A234600289).
| Metric | Measured Value |
|---|---|
| FPS (hw_only) | 457.24 |
| FPS (streaming) | 457.24 |
| HW Latency | 2.209000 ms |
| Power (streaming avg) | 2.21800 W |
| Power (streaming max) | 2.22300 W |
| Power (idle) | 0.74979 W |
| Quantization | INT8 (post-training, DFC calibration) |
| DFC Compiler | 3.33.1 |
| HailoRT | 4.20.0 |
| Measured On | Hailo-8 M.2 (P/N: HM218B1C2FAE, S/N: HLDDM2A234600289) |
Deployment
Deployed as a single HEF binary. No ONNX runtime, TensorFlow Lite, or Python inference stack required at the edge.
| Target silicon | Hailo-8 |
| Measured on | Hailo-8 M.2 (P/N: HM218B1C2FAE, S/N: HLDDM2A234600289) |
| DFC compiler | 3.33.1 |
| Quantization | INT8 (post-training, production telemetry calibration) |
| Runtime | HailoRT (vendor runtime) |
| Edge platform | Raspberry Pi 5 + Hailo AI HAT+ (M.2 Key M) |
Deployment procedure
Copy the .hef binary to the target device. hailortcli run loads the HEF directly into the Hailo-8 dataflow engine over PCIe. Inference begins immediately with deterministic per-frame latency. No model conversion, graph optimization, or warmup phase required.
References
- Jewell, A. (2026). AxonML: A Pure-Rust Deep Learning Framework for Edge Inference. AutomataNexus LLC. Technical whitepaper.
- Hailo Technologies Ltd. (2024). Hailo Dataflow Compiler User Guide. DFC v3.33.1.
- Hailo Technologies Ltd. (2024). Hailo-8 Product Datasheet.
AutomataNexus LLC · Fort Wayne, Indiana · andrew.jewellsr@automatanexus.com
Andrew Jewell Sr. · ORCID 0009-0005-2158-7060
May 2026 · All rights reserved.