add torch (chute template imports it)

bump to dodge cached image name collision

add pro_6000 NodeSelector include (chutes platform requires)

onnxruntime-gpu deps

onnxruntime miner.py (was ultralytics .pt loader)

R18 ONNX FP16 (1280, mAP50 0.932)

R18 migration: drop .pt (subnet requires .onnx)

R18 miner.py

R18 weights (mAP50 0.932, +0.052 F1 vs R17 on test)

R17 miner.py docstring + logging

R17 weights (mAP50 0.928, 1280imgsz)

Remove stale weights.onnx — best.pt is the active model (under 30MB cap)

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

Delete __pycache__/miner.cpython-312.pyc with huggingface_hub

Delete best_r7.pt with huggingface_hub

Delete best.pt with huggingface_hub

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

scorevision: push artifact

initial commit