Albedo

Install the distillation package and dependencies from source

pip install -e .

Run all 15 pre-submission validator checks locally to verify architecture, params, tokenizer, quantization, and format compliance

python test_miner.py --model-repo your-username/your-model

Full local eval including KL scoring (requires GPU); compare against current king with --king-repo flag(optional)

python check_model.py --model-repo your-username/your-model --eval

Dry-run submission (validates everything without committing on-chain); fix any validation errors before proceeding

python miner.py --network finney --netuid 97 --wallet-name my_wallet --hotkey-name my_hotkey --model-repo your-username/your-distilled-model --dry-run

Commit your model to the subnet (PERMANENT — interactive confirmation required; one commitment per hotkey, forever)

python miner.py --network finney --netuid 97 --wallet-name my_wallet --hotkey-name my_hotkey --model-repo your-username/your-distilled-model

Install the distillation package and dependencies (requires: bittensor, torch, transformers, vllm, fastapi, etc.)

pip install .

Register a validator hotkey on subnet 97 via the standard Bittensor flow

btcli subnet register --netuid 97

Run the validator via the wrapper script; or use PM2: pm2 start scripts/run_validator.sh --name distil-validator

bash scripts/run_validator.sh

Run the GPU eval pod remotely (Lium setup); vLLM teacher generation + student HF logit extraction; teacher logits GPU-resident, early stopping enabled

python scripts/pod_eval_vllm.py