Tenzro
Tutorial — Run a node

Run a trainer node

Tenzro Train is a Rust protocol with a Python reference trainer (PyTorch FSDP2 + Hivemind). Trainers enroll for a run, fetch shards, push gradients, and earn TNZO once the syncer finalizes the round.
Level
Advanced
Time
~30 min
Prerequisites
Python 3.11, GPU (8GB+ VRAM), staked TNZO
Stack
Python · CLI
01

Install the trainer

The reference trainer ships as a Python package with optional extras for language, vision, and confidential modes.

pip install "tenzro-trainer[language,vision]"
02

Enroll for an open run

The CLI handles identity, attestation (for Confidential tier), and the JSON-RPC handshake with the syncer.

tenzro-trainer enroll \
  --run-id YOUR_RUN_ID \
  --rpc https://rpc.tenzro.network
03

Run the inner loop

The trainer drives H steps of inner SGD, computes the OuterGradient, and pushes it to the syncer.

tenzro-trainer run \
  --run-id YOUR_RUN_ID \
  --adapter language \
  --shard ./data/shard-007
04

Submit and finalize

Submit the gradient — the syncer accumulates submissions and finalizes the round when the witness committee reaches quorum.

tenzro-trainer submit-gradient --run-id YOUR_RUN_ID
tenzro train get-run --run-id YOUR_RUN_ID
Related
Open the docs →Browse all tutorials
← All tutorials