Moirai-MoE-1.0-R-Base

Moirai-MoE-1.0-R-Base is Salesforce's sparse-expert extension of the Moirai line, the larger counterpart to Moirai-MoE-1.0-R-Small. Instead of relying on a single dense network for all behaviors, the Moirai-MoE design routes tokens through specialized experts to improve parameter efficiency and specialization across heterogeneous series.

Architecturally it is a sparse mixture-of-experts decoder-only transformer with probabilistic output heads, holding around 0.94B stored parameters. Because routing activates only a subset of experts per token, inference stays cheaper than an equally sized dense alternative while still supporting multivariate zero-shot forecasting with calibrated quantiles. The checkpoint card is sparse; Salesforce's official Moirai-MoE materials describe large heterogeneous time-series pretraining for the setup rather than a separate narrow corpus for this exact checkpoint.