Align metadata with AI Portal schemas
Replace the mixed model/benchmark document with standalone artifact metadata and an accuracy-only vision-detection report. Performance is omitted because the available timing is not the schema-required median end-to-end measurement.
Ready for review.
Scope updated: this PR now differs from HF main only by:
- replacing
metadata.yamlwith AI Portal model metadata; - adding
benchmark/accuracy.yamlwith the currentcoco-128-by-sameasurement andArm Optimized Runtimesas the runtime name; and - deleting
runtime/arm_yolo26n_pi5_model/model/model.bin.
The README, repro.py, model note, and reference report are restored byte-for-byte from HF main. Both YAML documents validate against the current AI Portal schemas. metadata.yaml uses the deliberately nonexistent filename MODEL_FILE_NOT_DISTRIBUTED; no placeholder or converted weights are included.
Correction: model.bin is intentionally absent. metadata.yaml points to that nonexistent path temporarily while the catalogue contract is resolved. No placeholder or converted weights are included.
refs/pr/2 ref