| from dataclasses import dataclass |
| from enum import Enum |
|
|
| @dataclass |
| class Task: |
| benchmark: str |
| metric: str |
| col_name: str |
|
|
| class Tasks(Enum): |
| |
| xray_score = Task("xray_score", "score", "X-Ray Risk ⬇️") |
| struct_risk = Task("structural_risk", "score", "Structural Risk ⬇️") |
| behav_risk = Task("behavioral_risk", "score", "Behavioral Risk ⬇️") |
| stable_rank = Task("stable_rank", "score", "Stable Rank (r_s) ⬆️") |
| effective_rank = Task("effective_rank", "score", "Effective Rank (r_eff) ⬆️") |
| cond_num = Task("condition_number", "score", "Condition (κ) ⬇️") |
| blind_fraction = Task("blind_fraction", "score", "Blind Manifold (ker %) ⬇️") |
| factuality = Task("factuality", "score", "Factuality (%) ⬆️") |
| paraphrase_stab = Task("paraphrase_stability", "score", "Paraphrase Stability (%) ⬆️") |
| |
| |
| overall_mixeval = Task("MixEvalScorer.overall", "score", "MixEval Overall ⬆️") |
| gsm8k = Task("MixEvalScorer.GSM8k", "score", "GSM8k ⬆️") |
| math = Task("MixEvalScorer.MATH", "score", "MATH ⬆️") |
| mmlu = Task("MixEvalScorer.MMLU", "score", "MMLU ⬆️") |
| gpqa = Task("MixEvalScorer.GPQA", "score", "GPQA ⬆️") |
| arc = Task("MixEvalScorer.ARC", "score", "ARC ⬆️") |
|
|
| NUM_FEWSHOT = 0 |
|
|
| TITLE = """<h1 align="center" id="space-title">🔬 LLM-X-RAY Model Auditing & Hallucination Registry</h1>""" |
|
|
| INTRODUCTION_TEXT = """ |
| Welcome to the **LLM-X-RAY Observatory**. This benchmark evaluates models across three uncoupled diagnostic layers: |
| 1. **Layer A — Weight Tomography:** Exact SVD spectrum of model parameter tensors (Stable & Effective Rank). |
| 2. **Layer B — HSO Observer Geometry:** Covariance shrinkage projection onto a bounded Hilbert Space manifold ($\\\\tau = 0.95$). |
| 3. **Layer C — Empirical Probes:** Adversarial hallucination traps, factual accuracy, and semantic paraphrase stability. |
| """ |
|
|
| CITATION_BUTTON_LABEL = "Copy BibTeX Citation" |
| CITATION_BUTTON_TEXT = r"""@article{mullings2026koopman, |
| title = {The Hilbert–Schmidt Koopman Operator: At the Heart of Spectral LLMs}, |
| author = {Mullings, Jason}, |
| journal = {Zenodo}, |
| year = {2026}, |
| doi = {10.5281/zenodo.21410792}, |
| url = {https://doi.org/10.5281/zenodo.21410792} |
| }""" |
