Moirai 2.0: When Less Is More for Time Series Forecasting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Chenghao, Aksu, Taha, Liu, Juncheng, Liu, Xu, Yan, Hanshu, Pham, Quang, Savarese, Silvio, Sahoo, Doyen, Xiong, Caiming, Li, Junnan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917243228520448
author Liu, Chenghao
Aksu, Taha
Liu, Juncheng
Liu, Xu
Yan, Hanshu
Pham, Quang
Savarese, Silvio
Sahoo, Doyen
Xiong, Caiming
Li, Junnan
author_facet Liu, Chenghao
Aksu, Taha
Liu, Juncheng
Liu, Xu
Yan, Hanshu
Pham, Quang
Savarese, Silvio
Sahoo, Doyen
Xiong, Caiming
Li, Junnan
contents We introduce Moirai 2.0, a decoder-only time-series foundation model trained on a new corpus of 36M series. The model adopts quantile forecasting and multi-token prediction, improving both probabilistic accuracy and inference efficiency. On the Gift-Eval benchmark, it ranks among the top pretrained models while achieving a strong trade-off between accuracy, speed, and model size. Compared to Moirai 1.0, Moirai 2.0 replaces masked-encoder training, multi-patch inputs, and mixture-distribution outputs with a simpler decoder-only architecture, single patch, and quantile loss. Ablation studies isolate these changes -- showing that the decoder-only backbone along with recursive multi-quantile decoding contribute most to the gains. Additional experiments show that Moirai 2.0 outperforms larger models from the same family and exhibits robust domain-level results. In terms of efficiency and model size, Moirai 2.0 is twice as fast and thirty times smaller than its prior best version, Moirai 1.0-Large, while also performing better. Model performance plateaus with increasing parameter count and declines at longer horizons, motivating future work on data scaling and long-horizon modeling. We release code and evaluation details to support further research.
format Preprint
id arxiv_https___arxiv_org_abs_2511_11698
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Moirai 2.0: When Less Is More for Time Series Forecasting
Liu, Chenghao
Aksu, Taha
Liu, Juncheng
Liu, Xu
Yan, Hanshu
Pham, Quang
Savarese, Silvio
Sahoo, Doyen
Xiong, Caiming
Li, Junnan
Machine Learning
We introduce Moirai 2.0, a decoder-only time-series foundation model trained on a new corpus of 36M series. The model adopts quantile forecasting and multi-token prediction, improving both probabilistic accuracy and inference efficiency. On the Gift-Eval benchmark, it ranks among the top pretrained models while achieving a strong trade-off between accuracy, speed, and model size. Compared to Moirai 1.0, Moirai 2.0 replaces masked-encoder training, multi-patch inputs, and mixture-distribution outputs with a simpler decoder-only architecture, single patch, and quantile loss. Ablation studies isolate these changes -- showing that the decoder-only backbone along with recursive multi-quantile decoding contribute most to the gains. Additional experiments show that Moirai 2.0 outperforms larger models from the same family and exhibits robust domain-level results. In terms of efficiency and model size, Moirai 2.0 is twice as fast and thirty times smaller than its prior best version, Moirai 1.0-Large, while also performing better. Model performance plateaus with increasing parameter count and declines at longer horizons, motivating future work on data scaling and long-horizon modeling. We release code and evaluation details to support further research.
title Moirai 2.0: When Less Is More for Time Series Forecasting
topic Machine Learning
url https://arxiv.org/abs/2511.11698