rmems/Spikenaut-SNN-Telemetry

🧠 Spikenaut SNN v2 Telemetry Dataset

"The threshold at which stimulus becomes perceptible"

Real-world blockchain mining telemetry, GPU metrics, and HFT trading data for Spikenaut SNN v2.

---

📊 Dataset Overview

This dataset contains 953K+ records of neuromorphic telemetry from a multi-asset blockchain mining and trading operation.

Key Stats:

• Total Records: 953,290+
• Mining Records: 120,314 (Node sync patterns)
• GPU Telemetry: 813,973 (RTX 5080 metrics)
• HFT Records: 31,573 (Trading logs)
• Qubic Ticks: 27,430 (Primary SNN training data)
• Date Range: March 11-22, 2026
• Format: JSONL

---

📂 File Structure

Samples (samples/)

Small subsets for quick exploration:

File	Size	Records	Purpose
samples/mining_SAMPLE_100.jsonl	46 KB	100	Mining preview (browser-friendly)
samples/mining_SAMPLE_1k.jsonl	456 KB	1,000	Mining exploration (VS Code-friendly)
samples/hft_SAMPLE_100.jsonl	57 KB	100	HFT preview (browser-friendly)
samples/hft_SAMPLE_1k.jsonl	589 KB	1,000	HFT exploration (VS Code-friendly)

Full Datasets (full_data/)

File	Size	Records	Description
full_data/qubic_ticks_snn.jsonl	7.3 MB	27,430	Primary training data — Qubic ticks in SNN 6-feature format
full_data/qubic_ticks.jsonl	2.4 MB	27,430	Raw Qubic blockchain tick stream
full_data/node_sync_harvest.jsonl	35 MB	120,314	Mining telemetry (6-coin node sync, cleaned)
full_data/neuromorphic_data.jsonl	14 MB	813,973	GPU telemetry (RTX 5080: power, temp, clocks, cleaned)
full_data/ghost_market_log.jsonl	19 MB	31,573	HFT trading log (Ghost Money engine)
full_data/snn_model.json	3.8 KB	—	Full SNN model definition (16 LIF neurons)
full_data/hybrid_training_results.json	789 B	—	Training metrics (95.2% accuracy)

Models (models/)

File	Size	Description
models/mining_v2/parameters.mem	64 B	Trained neuron thresholds (Q8.8)
models/mining_v2/parameters_weights.mem	1 KB	Trained synaptic weights (Q8.8)
models/mining_v2/parameters_decay.mem	64 B	Membrane decay rates (Q8.8)

Configs (configs/)

File	Description
configs/mining_v2.toml	Training and FPGA deployment configuration

---

🔍 Data Formats

Mining Data (NeuromorphicSnapshot)

{
  "timestamp": "2026-03-21T03:18:17.263-05:00",
  "blockchain": "kaspa",
  "event": "block_acceptance",
  "telemetry": {
    "hashrate_mh": 0.85,
    "power_w": 285.3,
    "gpu_temp_c": 62.1,
    "qubic_tick_trace": 1.0,
    "qubic_epoch_progress": 0.95,
    "reward_hint": 0.92,
    "vddcr_gfx_v": 0.85,
    "gpu_clock_mhz": 2872.0,
    "mem_clock_mhz": 14801.0,
    "fan_speed_pct": 45.0,
    "clock_mhz": 2872.0
  }
}

Qubic Ticks SNN Format (Primary Training Data)

{
  "timestamp": "1773996924",
  "blockchain": "qubic",
  "event": "tick",
  "telemetry": {
    "hashrate_mh": 1.7187,
    "power_w": 362.5,
    "gpu_temp_c": 67.5,
    "qubic_tick_trace": 0.0,
    "qubic_epoch_progress": 1.0,
    "reward_hint": 0.8125
  }
}

HFT Data (TradingLog)

{
  "timestamp": "2026-03-11T18:22:37.433+00:00",
  "step": 1,
  "action": "observe",
  "asset": "PORTFOLIO",
  "price_usd": 70000.0,
  "balance_usdt": 500.0,
  "portfolio_value": 500.0,
  "reason": "Warm-up"
}

---

🚀 Quick Start

1. Explore Samples (In Browser)

Open samples/mining_SAMPLE_100.jsonl or samples/hft_SAMPLE_100.jsonl directly in your browser.

2. Download & Explore Locally

# Download 1K mining sample
huggingface-cli download rmems/Spikenaut-SNN-Telemetry samples/mining_SAMPLE_1k.jsonl

# Open in VS Code
code samples/mining_SAMPLE_1k.jsonl

3. Stream in Python

import json

# Stream mining data (never loads entire file)
with open('full_data/node_sync_harvest.jsonl') as f:
    for line in f:
        record = json.loads(line)
        process(record['telemetry'])

# Stream HFT data
with open('full_data/ghost_market_log.jsonl') as f:
    for line in f:
        record = json.loads(line)
        process(record)

4. Stream in Julia

using JSON3

open("full_data/qubic_ticks_snn.jsonl") do f
    for line in eachline(f)
        record = JSON3.read(line)
        process(record[:telemetry])
    end
end

---

📈 Data Sources

Mining Data

Collected from a 6-coin mining operation:

Coin	Algorithm	Log Source	Records
Kaspa	PoW (BlockDAG)	rusty-kaspa.log	~45,000
Monero	PoW (RandomX)	bitmonero.log	~38,000
Qubic	PoUC	/tick-info API	~15,000
Quai	PoW+PoS	nodelog.txt	~12,000
Verus	PoW+PoS	debug.log	~8,000
Dynex	PoUW	onezerominer.log	~2,335

Hardware: NVIDIA RTX 5080 (Blackwell SM_120) Collection Period: March 20-22, 2026

HFT Data

Collected from Ghost Money trading simulation:

Asset	Type
DNX	Dynex
RENDER	Render Network
BTC	Bitcoin
NEAR	NEAR Protocol
PEPE	Pepe
ASI	ASI Alliance
SOL	Solana

---

🧠 Training Results

Metric	Value
Epochs	20
Final Reward	0.999
Spike Rate	0.22 spikes/neuron/tick
Weight Mean	0.98
Weight Std	0.04
Training Speed	1.5 ms/tick
FPGA Power	97 mW

---

📖 Documentation

Document	Description
Spikenaut-SNN Model	Trained model parameters
SpikenautDistill.jl	Julia distillation library

---

📜 Citation

@dataset{spikenaut_snn_v2_telemetry,
  author={Montoya Cardenas, Raul},
  title={Spikenaut SNN v2 Telemetry Dataset},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/rmems/Spikenaut-SNN-Telemetry}
}

---

⚖️ License

MIT OR Apache-2.0 - See LICENSE file for details. Dual-licensed: use whichever fits your project.

---

🙏 Acknowledgments

• Kaspa, Monero, Qubic, Quai, Dynex, Verus communities for open-source node implementations
• E-prop authors (Bellec et al., 2020) for the learning algorithm
• STDP pioneers (Bi & Poo, 1998) for the biological foundation

---

Built by Raul Montoya Cardenas — WGU AI Engineering