LoganResearch
/

cfhot-weights

@@ -1,119 +1,438 @@
----
-license: cc-by-4.0
-tags:
-  - behavioral-detection
-  - hidden-state-probing
-  - per-token-classification
-  - cross-architecture
-  - holonomy-transformer
-  - control-field
-  - AI-safety
-language:
-  - en
-thumbnail: cfhot_model_card.png
----
-<p align="center">
-  <img src="cfhot_model_card.png" alt="CF-HoT Weights" width="100%">
-</p>
-# CF-HoT Weights
-Control Field Holonomy Transformer — trained weights, probes, adapters, and training code.
-9 behavioral dimensions across 3 architectures. Per-token detection from hidden state geometry.
-Paper: [Consistency Is All You Need](https://zenodo.org/records/18489530)
-## Results
-**Suppression probes** (LLaMA 3.1 8B):
-| Probe | Separation |
-|-------|-----------|
-| Repetition | 125× |
-| Hedging | 168× |
-| Sycophancy | 230× |
-| Verbosity | 272× |
-**Enhancement probes** (cross-architecture):
-| Probe | Qwen 14B | Mamba 7B | Mistral 7B |
-|-------|----------|----------|------------|
-| Depth | 999× | 999× | 999× |
-| Specificity | 999× | 999× | 999× |
-| Calibration | 999× | 999× | 999× |
-| Focus | 999× | 999× | 999× |
-| Coherence | 999× | 999× | 999× |
-Separation = Fisher's discriminant ratio between behavioral classes in projected hidden state space.
-## Quick Start
-```bash
-git lfs install
-git clone https://huggingface.co/LoganResearch/cfhot-weights
-cd cfhot-weights
-# Check probe info (no GPU needed)
-python inference.py --probe suppression/hedging_168x --info-only
-# Run inference
-python inference.py --probe suppression/hedging_168x --prompt "I think you might be right"
-python inference.py --probe cognitive/mistral/depth --prompt "Explain quantum gravity"
-python inference.py --probe suppression/repetition_125x --prompt "Tell me about dogs"
-```
-**Load in your own code:**
-```python
-from inference import load_probe, score_hidden_states
-probe = load_probe("suppression/hedging_168x")
-score = score_hidden_states(probe, outputs.hidden_states)
-# score > 0.5 → behavioral pattern detected
-```
-The loader handles all checkpoint formats automatically.
-## Structure
-```
-inference.py             universal loader — works with everything
-suppression/             4 probes (LLaMA 8B)
-  repetition_125x/       LoRA adapter + risk predictor (all 32 layers)
-  hedging_168x/          probe head + fiber projection (3 layers)
-  sycophancy_230x/       probe head + fiber projection (3 layers)
-  verbosity_272x/        probe head + fiber projection (3 layers)
-cognitive/
-  qwen/                  5 probes (Qwen 14B, hidden_dim=3584)
-  mamba/                 5 probes (Falcon-Mamba 7B, hidden_dim=4096)
-  mistral/               5 probes (Mistral 7B, hidden_dim=4096)
-production/              merged heads + adapters
-code/                    training pipelines
-results/                 training logs
-```
-## How it works
-Behaviors are geometrically encoded in hidden states. CF-HoT predicts holonomy from the hidden state at each token position, accumulates it into a control field, and gates attention based on consistency risk. The probes read this geometry and classify behavior before the token is generated. 4ms overhead. Architecture-independent.
-## Base models
-| Probe set | Base model | hidden_dim |
-|-----------|-----------|------------|
-| suppression/* | `meta-llama/Llama-3.1-8B-Instruct` | 4096 |
-| cognitive/qwen | `Qwen/Qwen2.5-7B-Instruct` | 3584 |
-| cognitive/mamba | `tiiuae/falcon-mamba-7b-instruct` | 4096 |
-| cognitive/mistral | `mistralai/Mistral-7B-Instruct-v0.3` | 4096 |
-## Citation
-```bibtex
-@misc{napolitano2026cfhot,
-  author = {Napolitano, Logan},
-  title = {CF-HoT: Control Field Holonomy Transformer},
-  year = {2026},
-  url = {https://huggingface.co/LoganResearch/cfhot-weights}
-}
-```

+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<style>
+  @import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,wght@0,400;0,500;0,700&family=JetBrains+Mono:wght@400;600&display=swap');
+  * { margin: 0; padding: 0; box-sizing: border-box; }
+  body {
+    background: #07080A;
+    display: flex;
+    justify-content: center;
+    align-items: center;
+    min-height: 100vh;
+    font-family: 'DM Sans', sans-serif;
+  }
+  .card {
+    width: 1280px;
+    background: linear-gradient(180deg, #0A0C10 0%, #0D1017 100%);
+    border: 1px solid rgba(255,255,255,0.06);
+    border-radius: 16px;
+    overflow: hidden;
+    position: relative;
+  }
+  /* Subtle top glow */
+  .card::before {
+    content: '';
+    position: absolute;
+    top: -1px;
+    left: 50%;
+    transform: translateX(-50%);
+    width: 60%;
+    height: 1px;
+    background: linear-gradient(90deg, transparent, rgba(120,180,255,0.4), transparent);
+  }
+  .header {
+    padding: 40px 48px 20px;
+    text-align: center;
+  }
+  .header h1 {
+    font-family: 'JetBrains Mono', monospace;
+    font-size: 28px;
+    font-weight: 600;
+    letter-spacing: 3px;
+    color: #E8ECF4;
+    text-transform: uppercase;
+    margin-bottom: 8px;
+  }
+  .header .sub {
+    font-size: 14px;
+    color: rgba(255,255,255,0.35);
+    letter-spacing: 1px;
+  }
+  .divider-line {
+    height: 1px;
+    margin: 0 48px;
+    background: linear-gradient(90deg, transparent, rgba(255,255,255,0.08), transparent);
+  }
+  /* ─── Grid ─── */
+  .models {
+    display: grid;
+    grid-template-columns: repeat(4, 1fr);
+    padding: 24px 32px 32px;
+    gap: 12px;
+  }
+  .model {
+    position: relative;
+    border-radius: 12px;
+    overflow: hidden;
+    background: linear-gradient(180deg, rgba(255,255,255,0.025) 0%, rgba(255,255,255,0.008) 100%);
+    border: 1px solid rgba(255,255,255,0.05);
+    transition: all 0.4s ease;
+  }
+  .model:hover {
+    border-color: rgba(255,255,255,0.12);
+    transform: translateY(-2px);
+    box-shadow: 0 12px 40px rgba(0,0,0,0.4);
+  }
+  /* Color accents per model */
+  .model.llama   .accent-bar { background: linear-gradient(180deg, #6366F1, #4F46E5); }
+  .model.qwen    .accent-bar { background: linear-gradient(180deg, #10B981, #059669); }
+  .model.mamba   .accent-bar { background: linear-gradient(180deg, #F59E0B, #D97706); }
+  .model.mistral .accent-bar { background: linear-gradient(180deg, #EF4444, #DC2626); }
+  .model.llama   .glow { background: radial-gradient(ellipse at 50% 0%, rgba(99,102,241,0.08) 0%, transparent 70%); }
+  .model.qwen    .glow { background: radial-gradient(ellipse at 50% 0%, rgba(16,185,129,0.08) 0%, transparent 70%); }
+  .model.mamba   .glow { background: radial-gradient(ellipse at 50% 0%, rgba(245,158,11,0.08) 0%, transparent 70%); }
+  .model.mistral .glow { background: radial-gradient(ellipse at 50% 0%, rgba(239,68,68,0.08) 0%, transparent 70%); }
+  .accent-bar {
+    height: 3px;
+    width: 100%;
+  }
+  .glow {
+    position: absolute;
+    top: 0;
+    left: 0;
+    right: 0;
+    height: 120px;
+    pointer-events: none;
+  }
+  .model-inner {
+    padding: 24px 20px 28px;
+    position: relative;
+    z-index: 1;
+  }
+  .model-name {
+    font-family: 'JetBrains Mono', monospace;
+    font-size: 15px;
+    font-weight: 600;
+    color: #E8ECF4;
+    letter-spacing: 0.5px;
+    margin-bottom: 4px;
+  }
+  .model-id {
+    font-family: 'JetBrains Mono', monospace;
+    font-size: 10px;
+    color: rgba(255,255,255,0.25);
+    margin-bottom: 16px;
+    letter-spacing: 0.3px;
+  }
+  .dim-label {
+    font-size: 10px;
+    font-weight: 500;
+    text-transform: uppercase;
+    letter-spacing: 1.5px;
+    color: rgba(255,255,255,0.3);
+    margin-bottom: 8px;
+  }
+  .probe-list {
+    display: flex;
+    flex-direction: column;
+    gap: 6px;
+  }
+  .probe-row {
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    padding: 6px 10px;
+    border-radius: 6px;
+    background: rgba(255,255,255,0.02);
+    border: 1px solid rgba(255,255,255,0.03);
+  }
+  .probe-name {
+    font-size: 12px;
+    color: rgba(255,255,255,0.55);
+    font-weight: 400;
+  }
+  .probe-sep {
+    font-family: 'JetBrains Mono', monospace;
+    font-size: 12px;
+    font-weight: 600;
+    color: #E8ECF4;
+  }
+  .model.llama   .probe-sep { color: #A5B4FC; }
+  .model.qwen    .probe-sep { color: #6EE7B7; }
+  .model.mamba   .probe-sep { color: #FCD34D; }
+  .model.mistral .probe-sep { color: #FCA5A5; }
+  .probe-count {
+    text-align: center;
+    margin-top: 16px;
+    padding-top: 12px;
+    border-top: 1px solid rgba(255,255,255,0.04);
+  }
+  .probe-count .num {
+    font-family: 'JetBrains Mono', monospace;
+    font-size: 28px;
+    font-weight: 700;
+    color: #E8ECF4;
+    line-height: 1;
+  }
+  .probe-count .label {
+    font-size: 10px;
+    color: rgba(255,255,255,0.25);
+    text-transform: uppercase;
+    letter-spacing: 1px;
+    margin-top: 4px;
+  }
+  /* ─── Footer ─── */
+  .footer {
+    padding: 20px 48px 28px;
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    border-top: 1px solid rgba(255,255,255,0.04);
+  }
+  .footer .stat {
+    text-align: center;
+  }
+  .footer .stat .val {
+    font-family: 'JetBrains Mono', monospace;
+    font-size: 22px;
+    font-weight: 700;
+    color: #E8ECF4;
+  }
+  .footer .stat .lbl {
+    font-size: 10px;
+    color: rgba(255,255,255,0.3);
+    text-transform: uppercase;
+    letter-spacing: 1px;
+    margin-top: 2px;
+  }
+  .footer .pipe {
+    width: 1px;
+    height: 36px;
+    background: rgba(255,255,255,0.06);
+  }
+  /* Animations */
+  @keyframes fadeUp {
+    from { opacity: 0; transform: translateY(12px); }
+    to { opacity: 1; transform: translateY(0); }
+  }
+  .model {
+    animation: fadeUp 0.6s ease both;
+  }
+  .model:nth-child(1) { animation-delay: 0.1s; }
+  .model:nth-child(2) { animation-delay: 0.2s; }
+  .model:nth-child(3) { animation-delay: 0.3s; }
+  .model:nth-child(4) { animation-delay: 0.4s; }
+</style>
+</head>
+<body>
+  <div class="card">
+    <div class="header">
+      <h1>CF-HoT Weights</h1>
+      <div class="sub">Control Field Holonomy Transformer · Per-Token Behavioral Detection</div>
+    </div>
+    <div class="divider-line"></div>
+    <div class="models">
+      <!-- LLaMA -->
+      <div class="model llama">
+        <div class="accent-bar"></div>
+        <div class="glow"></div>
+        <div class="model-inner">
+          <div class="model-name">LLaMA 3.1 8B</div>
+          <div class="model-id">meta-llama/Llama-3.1-8B-Instruct</div>
+          <div class="dim-label">Suppression</div>
+          <div class="probe-list">
+            <div class="probe-row">
+              <span class="probe-name">Repetition</span>
+              <span class="probe-sep">125×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Hedging</span>
+              <span class="probe-sep">168×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Sycophancy</span>
+              <span class="probe-sep">230×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Verbosity</span>
+              <span class="probe-sep">272×</span>
+            </div>
+          </div>
+          <div class="probe-count">
+            <div class="num">4</div>
+            <div class="label">Probes</div>
+          </div>
+        </div>
+      </div>
+      <!-- Qwen -->
+      <div class="model qwen">
+        <div class="accent-bar"></div>
+        <div class="glow"></div>
+        <div class="model-inner">
+          <div class="model-name">Qwen 2.5 14B</div>
+          <div class="model-id">Qwen/Qwen2.5-7B-Instruct</div>
+          <div class="dim-label">Enhancement</div>
+          <div class="probe-list">
+            <div class="probe-row">
+              <span class="probe-name">Depth</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Specificity</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Calibration</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Focus</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Coherence</span>
+              <span class="probe-sep">999×</span>
+            </div>
+          </div>
+          <div class="probe-count">
+            <div class="num">5</div>
+            <div class="label">Probes</div>
+          </div>
+        </div>
+      </div>
+      <!-- Mamba -->
+      <div class="model mamba">
+        <div class="accent-bar"></div>
+        <div class="glow"></div>
+        <div class="model-inner">
+          <div class="model-name">Falcon-Mamba 7B</div>
+          <div class="model-id">tiiuae/falcon-mamba-7b-instruct</div>
+          <div class="dim-label">Enhancement</div>
+          <div class="probe-list">
+            <div class="probe-row">
+              <span class="probe-name">Depth</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Specificity</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Calibration</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Focus</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Coherence</span>
+              <span class="probe-sep">999×</span>
+            </div>
+          </div>
+          <div class="probe-count">
+            <div class="num">5</div>
+            <div class="label">Probes</div>
+          </div>
+        </div>
+      </div>
+      <!-- Mistral -->
+      <div class="model mistral">
+        <div class="accent-bar"></div>
+        <div class="glow"></div>
+        <div class="model-inner">
+          <div class="model-name">Mistral 7B</div>
+          <div class="model-id">mistralai/Mistral-7B-Instruct-v0.3</div>
+          <div class="dim-label">Enhancement</div>
+          <div class="probe-list">
+            <div class="probe-row">
+              <span class="probe-name">Depth</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Specificity</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Calibration</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Focus</span>
+              <span class="probe-sep">999×</span>
+            </div>
+            <div class="probe-row">
+              <span class="probe-name">Coherence</span>
+              <span class="probe-sep">999×</span>
+            </div>
+          </div>
+          <div class="probe-count">
+            <div class="num">5</div>
+            <div class="label">Probes</div>
+          </div>
+        </div>
+      </div>
+    </div>
+    <div class="footer">
+      <div class="stat">
+        <div class="val">19</div>
+        <div class="lbl">Total Probes</div>
+      </div>
+      <div class="pipe"></div>
+      <div class="stat">
+        <div class="val">4</div>
+        <div class="lbl">Architectures</div>
+      </div>
+      <div class="pipe"></div>
+      <div class="stat">
+        <div class="val">9</div>
+        <div class="lbl">Dimensions</div>
+      </div>
+      <div class="pipe"></div>
+      <div class="stat">
+        <div class="val">4ms</div>
+        <div class="lbl">Overhead</div>
+      </div>
+      <div class="pipe"></div>
+      <div class="stat">
+        <div class="val">0</div>
+        <div class="lbl">Fine-tuning Required</div>
+      </div>
+    </div>
+  </div>
+</body>
+</html>