Upload README.md with huggingface_hub

README.md

@@ -1,146 +1,52 @@
 ---
 language:
 - en
-license: gpl-3.0
-size_categories:
-- 1M<n<10M
+license: mit
 task_categories:
 - time-series-forecasting
-task_ids:
-- multivariate-time-series-forecasting
-pretty_name: Spikenaut SNN v2 — GPU Telemetry & Market Data
-tags:
-- spiking-neural-networks
-- neuromorphic
-- gpu-telemetry
-- cryptocurrency
-- blockchain
-- dynex
-- qubic
-- kaspa
-- monero
-- liquid-state-machine
-- stdp
-- julia
-- rust
-- cuda
+- feature-extraction
+pretty_name: "Spikenaut V2 Telemetry Data"
 ---
 
-# Spikenaut SNN v2 — GPU Telemetry & Market Data
+# Spikenaut V2 Telemetry Data
 
-**Author:** Raul Montoya Cardenas  
-**Institution:** Texas State University — Micro & Nano Device Systems (Spring 2026)  
-**Project:** [Eagle-Lander / Ship of Theseus](https://github.com/rmems/ship_of_theseus_rs)
+Multivariate time-series dataset for training a hardware-aware Spiking Neural Network (SNN) across 7 blockchain networks.
 
-Real hardware telemetry and live market data collected from an RTX 5080 mining rig running
-DNX, Qubic, Kaspa, and Monero workloads. Used to train and validate the Spikenaut V2
-4-lobe Ensemble Liquid State Machine (262,144 LIF neurons, CUDA-accelerated).
+## Dataset Description
 
----
+- **Total Events:** 1,132,861 validated neuromorphic events
+- **Format:** JSONL (JSON Lines)
+- **Pre-scrubbed:** All sensitive identifiers (wallets, IPs, hostnames) removed
 
-## Dataset Contents
+## Files
 
 | File | Rows | Size | Description |
-|------|------|------|-------------|
-| `data/neuromorphic_data.jsonl` | 789,114 | 363 MB | Primary GPU telemetry — voltage, temp, hashrate, power, clock, Qubic tick traces |
-| `data/ghost_market_log.jsonl` | 223,020 | 174 MB | Live market ticks — DNX, Qubic, Kaspa, Monero, Ocean, Verus price & volatility |
-| `data/node_sync_harvest.jsonl` | 120,334 | 55 MB | Blockchain node sync events — Kaspa, Monero, Dynex block acceptance rates |
-| `data/snn_model.json` | — | 3.8 KB | Trained SNN model weights snapshot |
-| `data/parameters.mem` | — | 80 B | FPGA-compatible membrane potential parameters (Q8.8 fixed-point) |
-| `data/parameters_decay.mem` | — | 80 B | FPGA-compatible decay parameters |
-| `data/parameters_weights.mem` | — | 1.3 KB | FPGA-compatible synaptic weight parameters |
-
-**Total: ~1.13M rows, 592 MB**
-
----
-
-## Telemetry Schema (`neuromorphic_data.jsonl`)
-
-Each row is a JSON object with a `telemetry` key:
-
-```json
-{
-  "telemetry": {
-    "vddcr_gfx_v": 0.7115,
-    "vram_temp_c": 59.0,
-    "gpu_temp_c": 51.0,
-    "hashrate_mh": 0.0,
-    "power_w": 156.9,
-    "gpu_clock_mhz": 2872.0,
-    "mem_clock_mhz": 14801.0,
-    "fan_speed_pct": 30.0,
-    "mem_util_pct": 27.0,
-    "power_z_score": 0.0,
-    "temp_z_score": 0.0,
-    "clock_z_score": 0.0,
-    "qubic_tick_trace": 0.935,
-    "qubic_tick_rate": 1.0,
-    "qubic_epoch_progress": 1.0,
-    "ocean_intel": 0.0,
-    "qu_price_usd": 0.0
-  }
-}
-```
-
----
-
-## Hardware
-
-- **GPU:** NVIDIA RTX 5080 (16 GB VRAM, Blackwell sm_120)
-- **Mining workloads:** DNX (Dynex), Qubic, Kaspa (KHeavyHash), Monero (RandomX)
-- **FPGA:** Artix-7 Basys3 — SNN reflex arc co-processor
-- **OS:** RHEL 8 / Fedora
-
----
-
-## SNN Architecture This Data Trains
-
-```
-MarketPulse (14 channels) + GPU Telemetry
-              ↓
-   [4 Parallel LIF Lobes]
-   ├── Scalper  τ=10ms  w=0.4
-   ├── Day      τ=25ms  w=0.3
-   ├── Swing    τ=50ms  w=0.2
-   └── Macro    τ=100ms w=0.1
-              ↓
-   Weighted Readout (16 channels)
-   Bull/Bear signals + confidence
-```
-
-- 262,144 total neurons (4 × 65,536)
-- Sparse recurrent connectivity (1%, ~42M synapses)
-- Ornstein-Uhlenbeck stochastic dynamics
-- STDP covariance learning
-- Hardware proprioception: GPU temp/power → global inhibition
-
----
-
-## Usage
-
-```python
-import json
-
-# Load GPU telemetry
-with open("data/neuromorphic_data.jsonl") as f:
-    for line in f:
-        row = json.loads(line)
-        t = row["telemetry"]
-        print(t["gpu_temp_c"], t["power_w"], t["qubic_tick_rate"])
-
-# Load market ticks
-with open("data/ghost_market_log.jsonl") as f:
-    for line in f:
-        tick = json.loads(line)
-        print(tick)
-```
-
----
-
-## License
-
-GPL-3.0 — same as the [Eagle-Lander](https://github.com/rmems/ship_of_theseus_rs) project.
-
----
-
-*Collected on ShipOfTheseus — Texas. All telemetry sanitized via the internal privacy scrubber.*
+|---|---|---|---|
+| `neuromorphic_data.jsonl` | 789,114 | 363 MB | GPU telemetry: voltage, temps, hashrate, power, clocks, fan speed, utilization |
+| `ghost_market_log.jsonl` | 223,020 | 174 MB | Live crypto market ticks: price & volatility for DNX, Qubic, Kaspa, Monero, Ocean, Verus |
+| `node_sync_harvest.jsonl` | 120,334 | 55 MB | Blockchain node sync events: block acceptance rates for Kaspa, Monero, Dynex |
+| `qubic_ticks.jsonl` | 27,430 | - | Qubic epoch & tick traces |
+| `snn_model.json` | - | 3.8 KB | Snapshot of trained SNN model weights |
+| `parameters.mem` | - | - | FPGA Q8.8 membrane potential parameters |
+| `parameters_decay.mem` | - | - | FPGA Q8.8 decay parameters |
+| `parameters_weights.mem` | - | - | FPGA Q8.8 synaptic weights |
+
+## Channels (V2: 16-channel)
+
+V2 processes 16 parallel channels across 4 trading lobes (Scalper, Day, Swing, Macro):
+- Hardware telemetry (GPU/CPU temp, power, clocks, fan)
+- Blockchain sync metrics (block acceptance, tick rates)
+- Market data (price, volatility across 6+ assets)
+- Qubic epoch/tick traces
+
+## Use Cases
+
+- Multivariate time-series forecasting for cryptocurrency markets
+- Hardware telemetry anomaly detection
+- FPGA co-processor deployment (Xilinx Artix-7)
+- SNN training with E-prop + OTTT learning rules
+
+## Related
+
+- **V2 Model:** [Spikenaut-SNN-v2](https://huggingface.co/rmems/Spikenaut-SNN-v2)
+- **V1 Dataset:** [Spikenaut-v1-Telemetry-Data](https://huggingface.co/datasets/rmems/Spikenaut-v1-Telemetry-Data)