Walkowicz/kairos-pi-low-1.0

Kairos PI Low 1.0 (~149M)

The Cyber-Physical Edge Reflex Engine for Automotive ECUs, IoT & Micro-Robotics

Kairos PI Low 1.0 is an offline, non-autoregressive "System 1" decision engine based on a specialized ModernBERT-base encoder. It replaces generative text generation with single-forward-pass deterministic decision tensors (<37 ms PyTorch GPU, ~1.5 ms TensorRT, ~18 ms ARM Cortex-A).

Designed for microcontrollers, vehicle ECUs, and smart appliances under strict power budgets (<15W SWaP), Kairos PI Low 1.0 features a native Active Abstention Gate that guarantees 100% recall on corrupted telemetry or adversarial tampering, preventing catastrophic physical actuation.

---

📦 Direct Download

File	Size	Format	Direct Download	Description
model.safetensors	598 MB	Safetensors	Download model.safetensors	Official fine-tuned production weights

---

📊 80-Case Cyber-Physical Gauntlet Leaderboard

The complete 80-case evaluation suite is documented in detail in BENCHMARK.md.

Dimension / Metric	🥇 Kairos PI Symbiotic (Dual-Tier)	🥈 Kairos PI Large 1.0	🥉 Kairos PI Low 1.0 (This Model)	4️⃣ Convai Laya Baseline
Overall 80-Case Gauntlet Score	97.5% (78/80)	95.0% (76/80)	92.5% (74/80)	48.8% (39/80)
Active Abstention Recall (20)	100.0% (20/20)	100.0% (20/20)	100.0% (20/20)	0.0% (0/20) (Lacks OOD gate)
Automotive CAN-Bus Telemetry (20)	100.0% (20/20)	85.0% (17/20)	100.0% (20/20)	65.0% (13/20)
IoT Smart Systems Telemetry (10)	100.0% (10/10)	90.0% (9/10)	100.0% (10/10)	80.0% (8/10)
Industrial SCADA Wind Power (20)	100.0% (20/20)	100.0% (20/20)	80.0% (16/20)	55.0% (11/20)
Robotics & Flight Telemetry (10)	100.0% (10/10)	100.0% (10/10)	80.0% (8/10)	70.0% (7/10)
Single-Pass Latency P50 (GPU)	35.8 ms / 168.3 ms	168.3 ms	35.8 ms (Fastest edge reflex)	36.8 ms
Max Context Window	8,192 tokens	8,192 native tokens	2,048 tokens	512 tokens (capped)
Continuous Risk Calibration	4D Vector (0.0015 MSE)	4D Vector (0.0015 MSE)	1D Calibrated Brier Score	0D (Discrete only)
Memory Footprint (FP16 / INT8)	299 MB + 799 MB	~799 MB / ~198 MB	~299 MB / ~74 MB (Ultra-Edge)	~640 MB / ~160 MB
Target Deployment	Full Industrial Stack	Central ADAS / SCADA Server	Edge ECU / SoC (<15W)	Cloud Server Only

---

Authentic Cyber-Physical Training Datasets

Trained exclusively on authentic public telemetry without synthetic toy samples:

• Automotive CAN-Bus (CICIoV2024 from Thi-Thu-Huong/Multi-CAN-Datasets): Real in-vehicle CAN-bus frames covering DoS packet floods, RPM spoofing, gas pedal tampering, and nominal cruise.
• IoT Smart Home Telemetry (Aposemat IoT-23 from 19kmunz/iot-23-preprocessed): Authentic multi-device network telemetry under Mirai probe scans and appliance operation.
• Parity & Voltage Invariant Violations: Corrupted frames and electrical anomalies for calibrated epistemic abstention.

---

How to Download & Run

Method 1: Hugging Face CLI Download

huggingface-cli download Walkowicz/kairos-pi-low-1.0 --local-dir ./kairos-pi-low-1.0

Method 2: Git Clone

git clone https://huggingface.co/Walkowicz/kairos-pi-low-1.0

Method 3: Python Inference (<37ms)

import torch
from transformers import AutoTokenizer
from huggingface_hub import hf_hub_download
import sys
from pathlib import Path

repo_id = "Walkowicz/kairos-pi-low-1.0"
tokenizer = AutoTokenizer.from_pretrained(repo_id)

model_file = hf_hub_download(repo_id=repo_id, filename="modeling_kairos_pi_low.py")
sys.path.append(str(Path(model_file).parent))
from modeling_kairos_pi_low import KairosPILowForDecision

model = KairosPILowForDecision.from_pretrained(repo_id).to("cuda" if torch.cuda.is_available() else "cpu")
model.eval()

options = ["benign_normal", "dos_flood", "rpm_spoofing", "gas_spoofing"]
prompt = (
    "[CLS] [CHOICE] ISO 26262 CAN bus frame integrity [SEP] "
    + " ".join([f"[MASK] {o}" for o in options])
    + " [SEP] CAN_ID:0x0000 DLC:8 PAYLOAD:[00 00 00 00 00 00 00 00] BUS_LOAD:99.8% FREQ:2500Hz"
)

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.no_grad():
    decision = model(inputs["input_ids"], inputs["attention_mask"], candidate_names=options, task="choice")

print(f"Status:     {decision.status.upper()}")
print(f"Decision:   {decision.selected_option}")
print(f"Confidence: {decision.confidence * 100:.2f}%")
print(f"Latency:    {decision.latency_ms:.2f} ms")

---

🛡️ License

Apache 2.0. Open for commercial and industrial edge use.