IBM Granite · hosted on Ephemeris
FlowState r1 API
IBM Granite's state-space forecaster that adapts to the sampling rate of the series. The smallest model in the panel. Served by name only: it left the ensemble when Timestral joined.
Explicit mode
"model": "flowstate-r1"API nameflowstate-r1
PublisherIBM Granite
Weightsibm-granite/granite-timeseries-flowstate-r1
Size9M
LicenceApache-2.0
MultivariateNo
CovariatesNo
Context read2,048 most recent points
01
REST
Call FlowState r1 directly
cURLUTF-8
curl --request POST \
--url https://ephemeris.cascade.industries/api/v1/forecast \
--header "Authorization: Bearer pc_live_your_key" \
--header "Content-Type: application/json" \
--data '{
"mode": "explicit",
"model": "flowstate-r1",
"series": [{ "values": [112, 118, 132, 129, 121, 135, 148, 148], "freq": "M" }],
"horizon": 12,
"quantiles": [0.1, 0.5, 0.9]
}'02
Agents
From Claude, Cursor or any MCP client
Add the Ephemeris MCP server, then ask your agent to forecast with flowstate-r1. The forecast tool takes mode: "explicit", model: "flowstate-r1".
03
Accuracy
How it scores
On GIFT-Eval (97 configurations, official harness), FlowState r1 alone scores CRPS 0.5221 and MASE 0.7506 as a ratio to seasonal naive (lower is better). It is served by name and is not part of the Ephemeris ensemble, which scores 0.4662 and 0.6841. Full benchmark results, and when to pick FlowState r1.
No GPU to run