Chronos-T5 (Large)

Chronos-T5 Large is the original large Chronos checkpoint, at 710M parameters. It is the reference model for the first Chronos generation: the largest of the classic tokenized checkpoints, kept available for teams that explicitly want the original autoregressive Chronos behavior rather than the newer designs.

It works by scaling and quantizing continuous time-series values into discrete tokens, then running a T5 encoder-decoder that autoregresses over future token sequences to recover probabilistic forecasts. This tokenize-then-autoregress approach is the defining trait of the first Chronos line. It is pretrained on the large public and synthetic time-series corpus used by the Chronos family, including Gaussian-process and KernelSynth-style synthetic augmentation described in the official Chronos paper.