LoganResearch
/

cfhot-weights

@@ -1,438 +1,127 @@
-<!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>

+---
+license: mit
+library_name: pytorch
+tags:
+- behavioral-detection
+- hidden-state-probing
+- per-token-classification
+- cross-architecture
+- holonomy-transformer
+- control-field
+- AI-safety
+- probes
+language:
+- en
+---
+<div align="center">
+<img src="cfhot_model_card.png" alt="CF-HoT Weights — 4 architectures, 19 probes" width="100%">
+</div>
+# 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
+pip install -r requirements.txt
+# 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
+# Load any probe — type and architecture auto-detected
+probe = load_probe("suppression/hedging_168x")
+# Score hidden states from any model forward pass
+score = score_hidden_states(probe, outputs.hidden_states)
+# score > 0.5 = behavioral pattern detected
+```
+The loader handles all checkpoint formats automatically:
+- Suppression probes (separate head + fiber_proj files)
+- Cognitive probes (single checkpoint with metadata)
+- Risk predictor (all-layer repetition detector)
+## 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}
+}
+```