File size: 8,448 Bytes
9bf4b1b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
{
  "model_id": "Metis-OLMoE-Latent-Telemetry",
  "base_model": "OLMoE-1B-7B-0125-Instruct-GGUF",
  "base_model_source": "allenai/OLMoE-1B-7B-0125-Instruct-GGUF",
  "version": "1.0.0",
  "license": "gpl-3.0",
  "repository_url": "https://huggingface.co/datasets/rmems/Metis-OLMoE-Latent-Telemetry",
  "description": "Spikenaut SNN Routing dataset containing raw bare-metal telemetry logs and latent space visualizations for SNN-quantized OLMoE MoE model",
  "research_purpose": "Map physical routing of LLM embeddings as processed by biologically-inspired neuronal fatigue mechanics",
  "primary_discovery": "Semantic Attractor Clustering - SNN physically routes different semantic concepts into distinct, repeatable biological pathways",
  
  "model_lineage": {
    "origin": {
      "model": "OLMoE-1B-7B-0125-Instruct",
      "organization": "AllenAI",
      "architecture": "Mixture of Experts (MoE)",
      "parameters": "1B active / 7B total",
      "quantization": "GGUF",
      "link": "https://huggingface.co/allenai/OLMoE-1B-7B-0125-Instruct-GGUF"
    },
    "derivation_pipeline": {
      "name": "corinth-canal",
      "purpose": "SNN quantization pipeline",
      "repository": "https://github.com/Limen-Neural/corinth-canal",
      "technique": "SAAQ (Semantic Attractor Architecture Quantization)"
    },
    "analysis_tools": {
      "visualization": {
        "name": "Surrogate_Viz.jl",
        "repository": "https://github.com/Spikenaut/Surrogate_Viz.jl",
        "purpose": "Symbolic regression and latent space visualization"
      }
    }
  },

  "experiment_chronology": [
    {
      "phase": 1,
      "name": "Synthetic Baseline (Smoke Test)",
      "directory": "origin_hardware_baselines/",
      "input": "Synthetic sine wave",
      "result": "Verified GPU temporal loop (10,000 ticks) and basic biological fatigue without crashing CUDA context",
      "hardware_baseline": "Resident Evil 4 Remake path tracing telemetry",
      "files": ["RE4_path_tracing_telemetry.csv"],
      "discovery": "Gaming workloads create dynamic 'heartbeat' vs static crypto-mining data"
    },
    {
      "phase": 2,
      "name": "The F16 Magnitude Collapse (Unbounded)",
      "directory": "first-day-testing-real-weights/first-test-falied/",
      "input": "Real LLM embeddings (OLMoE)",
      "issue": "Misconfigured CUDA kernels searching for F32 instead of F16",
      "result": "Routing collapse - unbounded electrical pressure caused single walker (~620) to become a 'blackhole'",
      "files": ["latent_space_exploration_first_real_attempt.png"],
      "discovery": "Raw F16-to-F32 extraction without scaling causes routing collapse"
    },
    {
      "phase": 3,
      "name": "Attractor Discovery (L2 Normalization)",
      "directory": "first-day-testing-real-weights/second-test/",
      "input": "Teaching OMLoE the language of SNN",
      "fix": "L2 Normalization applied to voltage",
      "result": "Energy settled into Walker 2000 with secondary echoes at Walkers 700 and 1450",
      "files": ["map_olmoe_english_logic.png"],
      "discovery": "L2 Normalization prevents Winner-Take-All collapse by bounding semantic pressure to unit sphere"
    },
    {
      "phase": 4,
      "name": "Rust Syntax (The True Victory)",
      "directory": "first-day-testing-real-weights/third-test/",
      "input": "fn main () { println!(); }",
      "result": "Routing changed completely from English prompt - code syntax routed to different biological neighborhood",
      "files": [
        "map_olmoe_rust_syntax_logic.png",
        "map_olmoe_rust_syntax_logic.txt",
        "snn_latent_telemetry.csv"
      ],
      "discovery": "L2 Normalization forces hardware to dynamically adapt to data type - different semantic concepts = different pathways",
      "walker_range": "600-800 frequency band",
      "key_metrics": {
        "ticks": 10000,
        "best_walker_range": "129-816",
        "elapsed_us_range": "206-4445"
      }
    },
    {
      "phase": 5,
      "name": "Math Logic Clustering",
      "directory": "first-day-testing-real-weights/fourth-test/",
      "input": "The derivative of a constant is mathematically zero.",
      "result": "Mathematical logic routed to exact same 600-800 frequency band as Rust syntax",
      "files": [
        "map_olmoe_math_logic.png",
        "telemetry_olmoe_math_logic.txt"
      ],
      "discovery": "Semantic Attractor Clustering - SNN maps highly structured logic tasks (math and code) to adjacent biological neighborhoods to conserve energy"
    }
  ],

  "key_concepts": {
    "walker": {
      "definition": "Pulse of electrical energy (spike) that physically explores network to find path of least resistance",
      "analogy": "Electrical impulses in biological brain"
    },
    "l2_normalization": {
      "purpose": "Prevents any single neuron from becoming dominant",
      "effect": "Mimics biological brain's energy distribution",
      "mathematical_result": "Bounds semantic pressure to unit sphere"
    },
    "semantic_attractor_clustering": {
      "definition": "SNN physically maps different semantic concepts to distinct, repeatable biological pathways",
      "example": "Abstract philosophy (2000-route) vs rigid code syntax (600-800 band)"
    },
    "fatigue_mechanics": {
      "definition": "Neurons that fire too much become less responsive",
      "purpose": "Prevents energy overload and enables network adaptation"
    }
  },

  "hardware_environment": {
    "workstation": "Ship of Theseus",
    "gpu": "ASUS ProArt GeForce RTX 5080 (16GB VRAM)",
    "cpu": "AMD Ryzen 9 9950X",
    "os": "Fedora 43",
    "implementation": "Custom Rust/CUDA corinth-canal"
  },

  "data_structure": {
    "routing": "CSV files containing routing and latent telemetry data",
    "experiments": "Test configurations and variants",
    "results": {
      "plots": "Visualization of SNN routing paths and firing density",
      "raw_telemetry": "Original tick-by-tick log files"
    }
  },

  "file_manifest": {
    "first-day-testing-real-weights/first-test-falied/": {
      "latent_space_exploration_first_real_attempt.png": "Visualization of routing collapse (blackhole at walker ~620)"
    },
    "first-day-testing-real-weights/second-test/": {
      "map_olmoe_english_logic.png": "English text semantic routing through Walker 2000"
    },
    "first-day-testing-real-weights/third-test/": {
      "map_olmoe_rust_syntax_logic.png": "Rust code syntax routing visualization (600-800 band)",
      "map_olmoe_rust_syntax_logic.txt": "Raw tick data for Rust syntax test (10,000 ticks)",
      "snn_latent_telemetry.csv": "CSV telemetry export for Rust syntax test"
    },
    "first-day-testing-real-weights/fourth-test/": {
      "map_olmoe_math_logic.png": "Mathematical logic routing visualization (600-800 band)",
      "telemetry_olmoe_math_logic.txt": "Raw tick data for math logic test (10,000 ticks)"
    },
    "origin_hardware_baselines/": {
      "RE4_path_tracing_telemetry.csv": "Resident Evil 4 thermal telemetry that inspired SAAQ equations"
    }
  },

  "usage": {
    "primary": "Feed SymbolicRegression.jl to discover new equations for SNN quantization",
    "secondary": "Train pure native Spikenaut SNN",
    "visualization": "Study spiking behavior, routing stability, and adaptive quantization"
  },

  "related_repositories": [
    {
      "name": "corinth-canal",
      "url": "https://github.com/Limen-Neural/corinth-canal",
      "purpose": "SNN quantization pipeline"
    },
    {
      "name": "Surrogate_Viz.jl",
      "url": "https://github.com/Spikenaut/Surrogate_Viz.jl",
      "purpose": "Symbolic regression and visualization"
    }
  ],

  "citation": {
    "bibtex": "@dataset{metis_olmoe_2025,\n  author = {Spikenaut},\n  title = {Metis-OLMoE-Latent-Telemetry: Spikenaut SNN Routing},\n  year = {2025},\n  publisher = {Hugging Face},\n  howpublished = {\\url{https://huggingface.co/datasets/rmems/Metis-OLMoE-Latent-Telemetry}}\n}",
    "apa": "Spikenaut. (2025). Metis-OLMoE-Latent-Telemetry: Spikenaut SNN Routing [Dataset]. Hugging Face. https://huggingface.co/datasets/rmems/Metis-OLMoE-Latent-Telemetry"
  },

  "tags": [
    "snn",
    "spiking-neural-network",
    "olmoe",
    "mixture-of-experts",
    "quantization",
    "neuromorphic",
    "telemetry",
    "semantic-routing",
    "latent-space",
    "cuda",
    "rust",
    "gguf"
  ],

  "created_at": "2025-04-16",
  "updated_at": "2025-04-16"
}