update documents

README.md

@@ -1,21 +1,31 @@
-# LiteRT models for Android dataset production
+---
+library_name: tensorflow-lite
+tags:
+- litert
+- tflite
+- android
+- on-device-training
+- computer-vision
+- model-collection
+---
 
-**Français / English** · Charlbi · Documentation 2026-09-21
+# LiteRT Models for Android Dataset Production
 
-25 variantes actives, dont 15 avec apprentissage de tête/adaptation. Poids téléchargés séparément de l’APK. Les encodeurs visuels restent figés.
+A collection of **25 current LiteRT conversion variants**, including **15 trainable
+heads or output adapters**, for experimental Android dataset and annotation workflows.
+Weights are downloaded separately from the application package. Visual backbones
+remain frozen; the trainable scope is defined by each variant's runtime contract.
 
-25 current variants, including 15 trainable heads/adapters. Model weights are downloaded separately from the APK. Visual backbones remain frozen.
+Maintained by [Unicorn Who Dev](https://huggingface.co/unicornwhodev).
+[English usage guide](docs/USAGE.en.md) · [Detailed results](docs/BENCHMARKS.en.md).
 
+## Recorded qualification
 
-## Résultats illustrés / Visual evidence
+The retained application campaign records **5 PASS, 1 FAIL and 19 PENDING** variants.
+All 15 learning variants passed synthetic host checks. RepViT M1 learning has a
+separate, report-level Android SDK result; it is not an application pass.
 
-[Résultats complets FR](docs/BENCHMARKS.md) · [Full results EN](docs/BENCHMARKS.en.md) · [Utilisation FR](docs/USAGE.md) · [Usage EN](docs/USAGE.en.md)
-
-![Application qualification](docs/figures/android-coverage.png)
-
-**5 succès applicatifs, 1 échec RTMDet, 19 variantes à tester.** / **5 application passes, 1 RTMDet failure, 19 variants pending.** A separate RepViT learning SDK report is documented separately. These are functional checks, not phone speed or accuracy benchmarks.
-
-## Current catalogue / Catalogue actif
+![Recorded Android application coverage](docs/figures/android-coverage.png)
 
 | Conversion | Host train | Standalone SDK | App |
 |---|---|---|---|
@@ -45,18 +55,65 @@
 | [vitpose](models/vitpose/README.md) | N/A | PENDING | PENDING |
 | [vitpose_learning](models/vitpose_learning/README.md) | PASS | PENDING | PENDING |
 
-## Reading the evidence / Lire les preuves
+PASS applies only to the indicated check. PENDING means unexecuted or unfinished;
+N/A means no training signature is exposed. The inference-only RTMDet Tiny failure
+is a feature-map/stride mismatch. Its learning variant is a different graph and
+passed its recorded application check.
+
+## What the checks establish
+
+The application checks used an **AOSP API 28 x86_64 software emulator**, LiteRT
+1.4.2, Select TF Ops 2.16.1 and a synthetic image. They establish execution and,
+where tested, optimizer updates, checkpoint restoration and resume. Whole-test
+durations include repeated inference and checkpoint operations and are not
+inference latency or FPS. No successful case produced a proposal on that fixture.
+
+Physical ARM device qualification, representative task accuracy, generalization,
+forgetting, memory, thermal behavior and repeated latency measurements remain open.
+These historical receipts do not qualify a later application binary.
+
+![Recorded checkpoint checks](docs/figures/android-checkpoints.png)
+
+[Successful application receipts](docs/evidence/android-app-passed.json)
+· [Failure receipt](docs/evidence/android-app-failed.json)
+· [Figure source data](docs/evidence/visual-data.json).
+
+## Download and integrate
+
+Read the selected `models/` folder's README, runtime contract and notices first.
+Download a pinned revision, verify its artifact hashes, and follow its exact
+preprocessing, tensor layout, labels and coordinate transform. Keep the original
+weights and save learned checkpoints separately. Evaluate a candidate on authorized,
+representative data before activation.
+
+```python
+from huggingface_hub import hf_hub_download
 
-![Android checkpoint proof](docs/figures/android-checkpoints.png)
+contract = hf_hub_download(
+    repo_id="unicornwhodev/Lite_rt_prepared_for_android_dataset_builder",
+    revision="36026262693de56b2cf45a6337a405297bfcfff6",
+    filename="models/edgenext_xx_small_learning/runtime_contract.json",
+)
+```
 
-The reports distinguish host conversion, standalone Android SDK and Vision Dataset Studio integration. Historical converter flags are retained in the original JSON; dated model-card updates link to later app receipts. No physical ARM phone, task accuracy or generalization is claimed.
+This example retrieves a contract; inference and learning use the model-specific
+instructions. Earlier fixed-shape variants are retained in
+[the historical archive](archive/static-20260919/README.md) and are excluded from
+the 25 current variants.
 
-Les rapports distinguent conversion hôte, SDK Android autonome et intégration à Vision Dataset Studio. Les indicateurs historiques des JSON restent inchangés ; les cartes datées renvoient aux reçus applicatifs ultérieurs. Aucune précision métier, généralisation ni qualification sur téléphone ARM n’est annoncée.
+## Rights and intended use
 
-## Files and rights / Fichiers et droits
+Each upstream model and component retains its own licence and notices. This
+collection grants no blanket licence to third-party weights. Preserve provenance
+and respect each model's permitted uses before download, adaptation or redistribution.
 
-Each `models/` folder preserves weights, hash manifests, runtime contracts, conversion/training reports and upstream notices where supplied. Historical fixed-shape versions are in [archive](archive/static-20260919/README.md). Use a pinned revision and verify hashes before loading. Each upstream licence applies independently; this collection does not grant a blanket licence to third-party weights.
+Application reference: [Vision Dataset Studio](https://github.com/unicornwhodev/vision-dataset-studio).
 
-Chaque dossier conserve poids, empreintes, contrats, rapports et attributions disponibles. Utiliser une révision épinglée, vérifier les hashes et consulter les conditions propres au modèle ; aucune licence globale des poids tiers n’est accordée ici.
+## Documentation and revision
 
-Application: [Vision Dataset Studio](https://github.com/unicornwhodev/vision-dataset-studio) · [Reproducible figure data](docs/evidence/visual-data.json).
+Card updated in English on **5 October 2026**, from the public repository at
+[`36026262693de56b2cf45a6337a405297bfcfff6`](https://huggingface.co/unicornwhodev/Lite_rt_prepared_for_android_dataset_builder/tree/36026262693de56b2cf45a6337a405297bfcfff6) and its retained reports.
+This update checks documentation, repository metadata and small evidence files;
+it does not rerun training, inference, dataset payload verification or device qualification.
+Historical receipts keep their original dates, revisions and scope. Earlier README
+hashes in artifact manifests refer to those earlier releases.