EiMon724/ScoreVision

My TurboVision Miner (local workspace)

This folder is your miner package (code + weights) that will be uploaded to your Hugging Face model repo and deployed to Chutes.

What you edit

• miner.py: your main engine (frame batch → boxes + keypoints)
• chute_config.yml: hardware + pip installs for your Chutes runtime
• Model files: put your weights/checkpoints in this folder too (e.g. *.pt, *.onnx, *.safetensors)

The contract (must match)

Your miner.py must contain:

• class Miner
• __init__(self, path_hf_repo: Path) to load weights from this folder
• predict_batch(self, batch_images, offset, n_keypoints) -> list[TVFrameResult]

Each returned TVFrameResult must include:

• frame_id: int
• boxes: list[{x1,y1,x2,y2,cls_id,conf}]
• keypoints: list[(x,y)] of length n_keypoints (pad missing with (0,0))

Deploy (from the turbovision repo root)

1. Ensure .env has your keys:

   • HUGGINGFACE_USERNAME, HUGGINGFACE_API_KEY
   • CHUTES_USERNAME, CHUTES_API_KEY
   • (optional for on-chain) BITTENSOR_WALLET_COLD, BITTENSOR_WALLET_HOT

2. Upload + deploy + (optionally) commit on-chain:

cd /home/pudge/Desktop/Score/turbovision
source .venv/bin/activate
sv -vv push --model-path /home/pudge/Desktop/Score/turbovision/my_miner_repo

First smoke test (after deploy)

Hit your Chutes /health and /predict endpoints (you’ll get the chute slug in deploy logs).