Home ／ Docs-Technical WhitePaper ／ 44-EFT.WP.Data.ModelCards v1.0
I. Chapter Purpose & Scope
Provide the layered taxonomy of Model Card fields (Required / Conditionally Required / Optional), naming & type constraints, minimal regex, and examples. All keys use snake_case; arrays denote plural entities with []; cross-volume citations use “Volume vX.Y:Anchor”.II. Field Layers & Naming Conventions
- Layers:
- Required: Must exist at release and pass type/regex/dependency checks.
- Conditionally Required: Must exist when trigger conditions apply (see “Trigger” column).
- Optional: Validate if present; absence is not an error.
- Naming & citations: Keys use snake_case. Fields that carry references use see[] with "Volume vX.Y:Anchor"; exports include references[] and version.
III. Required Fields
Key | Type | Constraint/Regex | Semantic Definition | Dependencies |
|---|---|---|---|---|
model_id | string | ^[a-z0-9_\\-.]+$ | Unique model identifier (root key for public release & lineage) | Export org per DataSpec. |
title | string | length ≥ 3 | Human-readable title | — |
version | string | ^v\\d+\\.\\d+(\\.\\d+)?$ | Semantic version (public cites use stable line) | Citation Spec v0.1. |
task | string | enum | Task type (classification/retrieval/generation/asr/timeseries, etc.) | Align with eval protocol. |
io_schema | object | schema | Input/Output modes (shape/dtype/constraints) | Consistent with deployment. |
architecture | object | schema | Architecture & parameters (backbone/modules/param count) | — |
training_data | object | schema | Training data references & frozen-split mapping | Reference DatasetCards. |
preprocess | object | schema | Train/infer preprocessing pipeline; parameter lock | Consistent with metrology. |
optimization | object | schema | Objective and optimizer/scheduler | — |
hyperparams | object | schema | Key hyperparameters & search posture | — |
evaluation | object | schema | Evaluation protocol & metrics; significance | Frozen splits; see quality chapter. |
calibration | object | schema | Probability calibration and coverage metrics | Consistent with metrology/uncertainty. |
robustness | object | schema | Shift/adversarial settings and robustness report | — |
fairness | object | schema | Fairness axes and gap thresholds | — |
safety | object | schema | Usage limits / harm mitigation / safety audits | Align with compliance. |
deployment | object | schema | Deployment form/device/concurrency & latency | — |
resources | object | schema | Resource usage (M_param/FLOPs/T_inf/QPS) | Units via Metrology chapter. |
export_manifest | object | schema | Export manifest with version, references[], artifacts | Machine-readable cites. |
IV. Conditionally Required Fields
Key | Type | Trigger | Constraint/Regex | Description | Dependencies |
|---|---|---|---|---|---|
path_dependence | object | Path-dependent quantities (e.g., T_arr) | schema | Register delta_form, path="gamma(ell)", measure="d ell" | Equations v1.1; Metrology v1.0. |
privacy_preserving | object | DP / secure inference | schema | dp_epsilon, encrypted inference posture | Compliance/privacy. |
compression | object | Distillation / pruning / quantization | schema | Methods & accuracy impact | Align with quality/eval. |
explainability | object | Explainability provided | schema | Methods (Grad/IG/LRP…) & coverage | — |
V. Optional Fields
Key | Type | Constraint | Description |
|---|---|---|---|
audit_trails | object | artifact traceability | Audit logs, DAGs, report link keys |
license_constraints | object | enum/policy | License constraints & region limits |
notes | string | — | Non-normative notes |
VI. task and io_schema Fragments (Normative)
task: "classification" # classification|retrieval|generation|asr|segmentation|timeseries|...
io_schema:
inputs:
- {name:"image", shape:"(H,W,3)", dtype:"uint8", range:"[0,255]"}
outputs:
- {name:"probs", shape:"(K,)", dtype:"float32", range:"[0,1]", semantics:"softmax"}
batching: {mode:"static|dynamic", max_batch: 64}
streaming: {enabled:false}
(I/O modes must align with deployment interfaces and eval protocol; units/dimensions follow the Metrology chapter.)
VII. Training-Data Binding & Frozen Splits (Excerpt)
training_data:
refs:
- "EFT.WP.Data.DatasetCards v1.0:Ch.11" # splits & distribution
- "EFT.WP.Data.DatasetCards v1.0:Ch.12" # quality & baselines
splits_ref: "eift.obs.demo@v1.0" # frozen-split reference
contamination_policy: "forbid-cross-split"
(The Model Card references data facts—no duplication; references carry version and anchors.)
VIII. Evaluation & Calibration (Excerpt)
evaluation:
protocol: {splits:"frozen", seeds:[0,1,2,3,4], repeats:5,
significance:{test:"permutation", alpha:0.05}}
metrics: [{name:"f1_macro"}, {name:"roc_auc"}, {name:"ece"}, {name:"brier"}]
calibration:
method: "temperature"
params: {t: 1.7}
report: {ece: 0.018, brier: 0.072}
(Significance & coverage metrics align with quality/metrology chapters.)
IX. Export Manifest & References (export_manifest Fragment)
export_manifest:
version: "v1.0"
artifacts:
- {path:"model_card.yaml", sha256:"..."}
- {path:"eval/summary.csv", sha256:"..."}
references:
- "EFT.WP.Core.DataSpec v1.0:EXPORT"
- "EFT.WP.Core.Metrology v1.0:check_dim"
- "EFT.WP.Core.Equations v1.1:S20-1"
(Exports must include version and references[]; references carry Volume+Version+Anchor.)
X. Key Patterns & Minimal Regex
- IDs & versions: model_id: ^[a-z0-9_\\-.]+$; version: ^v\\d+\\.\\d+(\\.\\d+)?$.
- Reference anchors: references[*]: ^[^:]+ v\\d+\\.\\d+:[A-Z].+$.
- Dimensional checks: metrology.units="SI" and check_dim=true.
XI. see[] & Cross-Volume Mapping (Example)
see:
- "EFT.WP.Data.DatasetCards v1.0:Ch.11"
- "EFT.WP.Core.Metrology v1.0:check_dim"
- "EFT.WP.Core.Equations v1.1:S20-1"
(Prefer clause-level cites and P/S/M/I anchor classes where available.)
XII. Chapter Compliance Checklist
- All required keys present and passing type/regex checks; reserved names not repurposed.
- export_manifest includes version and references[]; see[] uses "Volume vX.Y:Anchor".
- Any T_arr entries register delta_form, path, measure, and pass check_dim.
- Math/symbols follow backtick and parenthesis rules with no Chinese; T_fil/T_trans and n/n_eff never mixed.
Copyright & License (CC BY 4.0)
Copyright: Unless otherwise noted, the copyright of “Energy Filament Theory” (text, charts, illustrations, symbols, and formulas) belongs to the author “Guanglin Tu”.
License: This work is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0). You may copy, redistribute, excerpt, adapt, and share for commercial or non‑commercial purposes with proper attribution.
Suggested attribution: Author: “Guanglin Tu”; Work: “Energy Filament Theory”; Source: energyfilament.org; License: CC BY 4.0.
First published： 2025-11-11｜Current version：v5.1
License link：https://creativecommons.org/licenses/by/4.0/