v3 weights (yolo11s, 204k merged plate dataset, val mAP50=0.944)

v3 weights (yolo11s, 204k merged plate dataset, val mAP50=0.944)

ep29 weights + miner.py pipeline tuning (NMS sigma 0.3, SAHI overlap 5%, no adaptive fallback)

TEE mandate: include=pro_6000, price 2.00, Chute.tee=true

revert NodeSelector to Apr-16 horse config (no include, price 0.50)

NodeSelector: add include=[pro_6000] for TEE requirement on integrated subnet chutes

cleanup: remove stray yolo26n.pt (unused by chute)

v8 ep-10 YOLO26s weights + shape-dispatch miner.py + defensive guards

v3 weights (yolo11s, 204k merged plate dataset, val mAP50=0.944)

v3 weights (yolo11s, 204k merged plate dataset, val mAP50=0.944)

v3 weights (yolo11s, 204k merged plate dataset, val mAP50=0.944)

v3 weights (yolo11s, 204k merged plate dataset, val mAP50=0.944)

chute_config: fix param names (shutdown_after_seconds, max_hourly_price_per_gpu moved to NodeSelector). Previously silently dropped → chute deprovisioned after 5min idle.

chute_config: trim exclude to 4 items (Chutes max=5); rely on max_hourly_price for cost filtering

chute_config: exclude list cleaned up — only valid Chutes GPU identifiers, allow 3090/4090 only

chute_config: shutdown_after=288000 (80h) per ops guide (ALWAYS 288000, never lower)

chute_config: shutdown_after=28800 (8h idle before chute shuts down)

chute_config: quote GPU exclude strings (YAML was parsing digits as int)

chute_config: GPU restricted to 4090/3090, max_hourly_price 0.50

Promote v3 epoch 44 weights + quad-4 miner.py to standalone repo (was pushed to shared ScoreVision by mistake).

numberplate: v2b weights + miner + chute_config (additive, no deletions)

numberplate: quad-4 inference + v3 epoch-19 FP16 weights. Standalone repo for 30MB compliance.

initial commit