Tutorial — Training & governance

Post a training task

Tenzro Train splits cleanly: the Rust protocol crate owns aggregation, witness committee, and on-chain receipts; the Python reference trainer wraps PyTorch FSDP2 + Hivemind for the inner loop. You post a task, trainers enroll, and a k-of-N committee finalizes each round.

Level: Advanced
Time: ~30 min
Prerequisites: Trainer node stake
Stack: CLI · Python

Post the task

Specify the architecture, dataset reference, rounds, and aggregation rule.

tenzro train post-task \
  --architecture timesfm-2.5 \
  --dataset-ref ipfs://Qm... \
  --rounds 50 \
  --aggregation mean \
  --trust-tier open

Enroll a trainer

Each trainer node submits its DID and (for Confidential tier) its enclave attestation.

tenzro train enroll-trainer --run-id $RUN

Submit outer gradients

The Python reference trainer runs the inner loop and submits a signed outer gradient.

tenzro-trainer run --run-id $RUN --rounds 50

Read the receipt

After the witness committee finalizes the final round, the run produces a receipt with the model commitment.

tenzro train get-receipt --run-id $RUN

Open the docs →Browse all tutorials

← All tutorials