Serve a model on Tenzro

The CLI uses hf-hub under the hood and verifies SHA-256 integrity.

tenzro model download qwen-3-0.6b

The local runtime exposes an OpenAI-compatible HTTP surface for development.

tenzro model serve qwen-3-0.6b --listen 127.0.0.1:11434

Bond a small stake, then publish the model endpoint, price, and SLA to the catalog.

tenzro provider register
tenzro model endpoint publish \
  --model qwen-3-0.6b \
  --price-per-token 0.0000025

Once the registration confirms, the inference router will start placing requests with you.

tenzro inference request qwen3-0.6b "what is the chain id?"

The router hedges against slow providers: if a primary has not answered by its observed p95 tail latency, the same request goes to a backup and the first answer wins. Read the live counters to see how your endpoint is faring — a rising hedges_dispatched or deadline_exceeded against you means you are the slow tail and requests are being raced away or abandoned.

tenzro inference router-metrics
# requests           total routed
# hedges_dispatched  primary still pending past the delay
# hedges_won         backup answered first
# deadline_exceeded  abandoned on the wall-clock deadline