Tutorial — Multi-modal AI

Forecast with TimesFM

TimesFM 2.5 is a 200M-parameter timeseries foundation model. Tenzro exposes it through the forecast runtime — point it at a context window and get back a horizon of point or quantile predictions.

Level: Beginner
Time: ~10 min
Prerequisites: Tenzro CLI installed
Stack: CLI · JSON-RPC

Load the model on a provider

TimesFM ships under the Permissive license tier — load once per provider. The ONNX graph has to be on the node's filesystem already; --catalog-id then supplies the context window, horizon ceiling and output tensor name from the catalog entry.

tenzro forecast catalog

tenzro forecast load \
  --model fc \
  --path /models/timesfm-2.5-200m.onnx \
  --catalog-id timesfm-2.5-200m

Prepare your input series

The history is a flat array of floats — one observation per step, oldest first. The CLI takes it inline as a comma-separated list, or from a file holding a JSON array.

cat > history.json <<'JSON'
[101.2,102.5,103.1,104.0,103.7,105.2,106.0,107.1]
JSON

Run a forecast

Point predictions come back by default. Adding --quantile returns prediction intervals instead of a single path.

tenzro forecast run --model fc --context-file history.json --horizon 64

tenzro forecast run \
  --model fc \
  --context-file history.json \
  --horizon 64 \
  --quantile 0.1,0.5,0.9

Call from JSON-RPC

The same runtime is reachable as a typed JSON-RPC method. The series is keyed history on the wire; quantiles and frequency_seconds are optional.

curl -s https://rpc.tenzro.xyz -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tenzro_forecast","params":{"model_id":"fc","history":[101.2,102.5,103.1],"horizon":32,"quantiles":[0.1,0.5,0.9]}}'

Open the docs →Browse all tutorials

← All tutorials