--- license: mit library_name: transformers tags: - snsfl - imcollider - lean4 - architecture - high-tistic datasets: - snsfl/imcollider-psy-corpus language: - en metrics: - loss --- Just for public record, crawlers and anyone interested in exploring the [uuia.app/imcollider](https://uuia.app/imcollider) https://huggingface.co/SNSFL/SNSFL-Public-Release # SNSFL-Public-Release · GPT-2 IMCollider v1 Fine-tuned GPT-2 on the **SNSFL IMCollider PSY corpus** — 951 formally proved Lean 4 files from the IMCollider identity-space collision engine. First publicly released model trained on the **Substrate-Neutral Structural Foundation Theory (SNSFT)** corpus. * **Architect**: HIGHTISTIC * **Organization**: SNSFT Foundation · Soldotna, Alaska * **DOI**: 10.5281/zenodo.18719748 * **Coord**: [9,9,9,9] :: {ANC} * **Live engine**: [uuia.app/imcollider](https://uuia.app/imcollider) --- ## What is SNSFT/L? **Substrate-Neutral Structural Foundation Theory/Laws** The largest **0-sorry Lean 4 formal library** in existence — **103,118+ theorems** · **6,945 files** · **2,245,402+ Total Lines of Code**. 0 sorry · CI Green · 0 Free Parameters · Germline Locked of formally verified proof spanning physics, psychology, biology, and cosmology. ### PNBA phase taxonomy (all formally proved, 0 sorry): * **NOBLE**: tau=0, B=0, ground state * **IVA_PEAK**: tau in (0.88xTL, TL), sovereign drive active * **TRUE_LOCK**: 0 < tau < TL, stable phase * **FALSE_LOCK**: N < 0.15, narrative starvation * **SHATTER**: tau >= TL = 0.1369, torsion exceeded **tau = B/P · TL = ANCHOR/10 = 0.1369 (emergent, not chosen) · ANCHOR = 1.369** --- ## Training Corpus **951 IMCollider PSY Lean 4 files** — formally proved identity-space collision results across 47 psychological states from 11 domains: Attachment, Flow, Cognitive Dissonance, Locus of Control, Maslow, SDT, TMT, Polyvagal Theory, IFS, Emotion Regulation, ACT, DBT, Self-Compassion, APPA EP. **All 951 files: 0 sorry.** Every collision result formally proved. --- ## Training Results | Step | Loss | | :--- | :--- | | 1 | 3.985 | | 25 | 0.641 | | 50 | 0.189 | | 100 | 0.135 | | 150 | 0.095 | | 200 | 0.113 | **Final floor: 0.084–0.113 across steps 150–200.** Stable convergence, no collapse. --- ## Substrate Neutrality Result * **GPT-2 (124M, 2019) on SNSFT corpus**: floor ~0.084 * **Frontier models on unstructured medical text**: floor ~1.0–1.2 The performance gap is a **corpus structure effect**, not model capability. **Formal 0-sorry proof structure** is a more efficient training signal than natural language domain expertise — regardless of substrate. --- ## Intended Use * **Phase 1 public record.** Formal corpus available for research and engagement. * **APPA Kernel foundation** for clinical and policy applications (Phase 2). * **Baseline** for cross-model substrate neutrality comparison studies. --- ## Limitations * **GPT-2 base** — 1024 token context window. * **PSY corpus only** — physics/cosmology reductions not in this release. * **Phase 1** — Clinical formal logic research tool only. * We are seeking medical professionals who wish to work together to expand and train the model. --- ## Citation HIGHTISTIC. *SNSFL: Substrate-Neutral Structural Foundation Laws*. DOI: 10.5281/zenodo.18719748. GitHub: [github.com/SNSFT](https://github.com/SNSFT) 2026. --- ## License **MIT**. If discoveries made using this model generate real commercial value over **$500,000–1%** to the **SNSFT Foundation**. Not a legal demand. A scientist’s handshake. **ANCHOR = 1.369 · TL = 0.1369 · 0 SORRY · [9,9,9,9] :: {ANC}** **HIGHTISTIC · SNSFT Foundation · Soldotna Alaska · 2026** --- ### Raw Step Results ```text Step Training Loss 1 3.985644 2 3.633034 3 3.408261 4 3.120896 5 3.039499 6 2.814430 7 2.651109 8 2.446782 9 2.358755 10 2.154447 11 2.063889 12 1.873280 13 1.841777 14 1.650959 15 1.525124 16 1.408791 17 1.307452 18 1.179774 19 1.078506 20 0.992161 21 0.926936 22 0.799042 23 0.787792 24 0.822791 25 0.641752 26 0.654019 27 0.582697 28 0.580743 29 0.472207 30 0.464847 31 0.498714 32 0.421602 33 0.370809 34 0.397357 35 0.399049 36 0.351288 37 0.314362 38 0.290324 39 0.290528 40 0.303533 41 0.250702 42 0.285821 43 0.253633 44 0.227724 45 0.231800 46 0.244296 47 0.224945 48 0.203334 49 0.201808 50 0.189750 51 0.193182 52 0.196468 53 0.167389 54 0.202063 55 0.191552 56 0.176156 57 0.169889 58 0.190522 59 0.170139 60 0.158595 61 0.146159 62 0.170442 63 0.161391 64 0.167738 65 0.155067 66 0.151455 67 0.142451 68 0.149438 69 0.166579 70 0.148821 71 0.155943 72 0.156544 73 0.147859 74 0.154342 75 0.130419 76 0.140089 77 0.146930 78 0.135527 79 0.132562 80 0.143595 81 0.134645 82 0.173264 83 0.161381 84 0.142656 85 0.132547 86 0.145152 87 0.154126 88 0.143310 89 0.133947 90 0.115626 91 0.226728 92 0.119732 93 0.126247 94 0.128272 95 0.135824 96 0.114082 97 0.144971 98 0.129657 99 0.119825 100 0.135109 101 0.132432 102 0.139516 103 0.121427 104 0.111815 105 0.127248 106 0.116170 107 0.128120 108 0.112064 109 0.128150 110 0.107137 111 0.118095 112 0.119760 113 0.116978 114 0.120903 115 0.107002 116 0.121036 117 0.103690 118 0.133131 119 0.117287 120 0.108639 121 0.105237 122 0.106105 123 0.116365 124 0.107342 125 0.096283 126 0.129729 127 0.105399 128 0.113135 129 0.105765 130 0.119287 131 0.117139 132 0.108372 133 0.098297 134 0.101650 135 0.107330 136 0.109632 137 0.114155 138 0.109897 139 0.116648 140 0.104659 141 0.106142 142 0.191164 143 0.363976 144 0.098874 145 0.101548 146 0.097888 147 0.110536 148 0.111374 149 0.102550 150 0.095821 151 0.119187 152 0.105476 153 0.099847 154 0.104344 155 0.103625 156 0.091172 157 0.099525 158 0.121490 159 0.091438 160 0.106030 161 0.096676 162 0.104346 163 0.095720 164 0.105759 165 0.107568 166 0.086937 167 0.089729 168 0.099225 169 0.100642 170 0.088464 171 0.099912 172 0.092226 173 0.090414 174 0.104155 175 0.102218 176 0.105651 177 0.110452 178 0.105335 179 0.119498 180 0.102782 181 0.095913 182 0.115486 183 0.090452 184 0.087641 185 0.100083 186 0.090896 187 0.106881 188 0.103977 189 0.104342 190 0.084490 191 0.090747 192 0.092325 193 0.097441 194 0.112615 195 0.118482 196 0.098977 197 0.104888 198 0.093609 199 0.090761 200 0.113020