Restore Ara240 throughput configurations

List every competing configuration; add the runtime axis

Rank leaderboards on the ranked metric, not on the smart decoder

Fix: iPhone sessions were being attributed to the M2 Max CoreML CPU row

Add Snapdragon Hexagon HTP rows; publish all decode variants

Speed up anchored tab paint: early bootstrap before JSON, defer Chart.js

Fix tab deep-links: no flash, sync parent URL via postMessage

Update social card metrics and add linkable tabs

Update social card metrics and add linkable tabs

Upload folder using huggingface_hub

Remove generic Linux aarch64 platform coverage row and hardware badge

Refresh dashboard: EDGEAI-1356 arm64 ONNX regression fixed (profiler 1.14.3)

Add AWS EC2 platforms to coverage/leaderboard; fix badge and project-link bugs

Refresh full AWS Batch sweep; exclude EDGEAI-1356 arm64 ONNX regression (48 sessions)

Add imx95-ezurio Neutron sweep results

Refresh dashboard: imx95-verdin Neutron config fix (CMA depth 8, smart-latency)

Add Apple iPhone 17/15 Pro CoreML platforms (EFPI iOS sweep, 126 sessions)

Fix legacy Kinara naming in ecosystem/lifecycle diagrams (NXP Ara240 per BRANDING.md)

Fix legacy Kinara naming in ecosystem/lifecycle diagrams (NXP Ara240 per BRANDING.md)

Fix legacy Kinara naming in ecosystem/lifecycle diagrams (NXP Ara240 per BRANDING.md)

Refresh model zoo snapshot (334 sessions, generated_at 2026-07-04T18:07:18Z)

Refresh dashboard from the validator-ng v8.19.0 full-zoo reprocess (334 configurations, corrected throughput/ceiling model)

Theoretical-throughput leaderboards + ceiling parity chart; audited data (334); Neutron yolo11/26 = In progress

Refresh social card: 4 · 7 · 10 · 300+

Refresh dashboard: onnx-cuda 1.8.1 references + updated session counts

Filter superseded session versions: on-target platforms now consistent on profiler 1.8.1 (356->334 sessions)

Ara240: NPU-core Theoretical FPS + full latency/throughput sweep; dedup fix

Report realized vs theoretical FPS; fix duplicate-row dedup

Report realized vs theoretical FPS; fix duplicate-row dedup

Apply EdgeFirst/NXP branding rules (strict NXP i.MX prefixing, Ara240, trademark notice)

Apply EdgeFirst/NXP branding rules (strict NXP i.MX prefixing, Ara240, trademark notice)

Update Model Zoo dashboard: Apple CoreML results (13/21 models complete)

Update Model Zoo dashboard: ara240 yolo11/26 now shown, refreshed validation results

Refresh: latest Studio results (220 records)

Refresh: latest Studio results (211 records); failed-run guard active

Refresh: latest Studio results (208 records); broken Neutron 3.1.2 batch removed

Dedup re-runs by newest: Neutron re-runs now reflected

Refresh dashboard: 197 validation sessions (FPS distribution, decode strategy, reference provenance)

Capture EP/runner from description; Ara240 INT16; latest results

Prefer CUDA for the FP32 reference; refresh from latest snapshot

Label ONNX CPU benchmarks by chip family (Graviton/Intel/RPi5...)

Distinguish ONNX FP32/FP16 (CUDA/CPU) precisions; correct labels + dedup

Dedup genuine re-run duplicates in dashboard count (143→111); keep decoder variants + CoreML backends

Refresh dashboard from saas snapshot (2026-06-17) + hardened publish pipeline

Add hardware badges with chart-consistent colors to README

Add aarch64 to PLATFORM_ROWS; hardware badge colors updated

Change ARA240 chart color to light blue (#5BB8F5)

Refresh snapshot 2026-06-17; ARA240 regex fix fully live

Fix ARA240 sessions routing to ara240 platform row; mark unsupported for yolo11/yolo26

Update Supported Hardware: add x86_64/macOS, ARA240, i.MX 95 caveat