mlsanityData to 7 Oct 2026

Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting

arXiv 2605.29401 ↗First posted 28 May 2026cs.LGcited by 2

On the map: LLM reasoning → Time series reasoning

Abstract (arXiv)

Time-Series Foundation Models (TSFMs) excel at zero-shot unimodal forecasting using numerical data, but unlike LLMs they cannot consume multimodal, non-numerical context that often shape real-world trajectories. In this work, we bridge this gap and argue for a multimodal time-series forecasting approach that post-trains LLMs to act as context-guided revisors over strong numerical TSFM priors. We introduce PostTime, a post-training recipe combining Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Rewards (RLVR), along with a methodology to generate automated reasoning traces for forecast revisions. PostTime teaches an LLM to generate context-conditioned forecast interventions -- decisions to revise, preserve, or ignore the TSFM prior based on the multimodal context. We evaluate this approach on the TimesX multimodal forecasting benchmark using a Gemma-3-4B LLM and TimesFM-2.5 TSFM, and show that it significantly outperforms standalone TSFMs, LLM-only baselines, and existing multimodal forecasting approaches.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Time series reasoning, one of 17 directions in LLM reasoning. Builds on 25 of its 28 references that are in the map. First posted on arXiv 28 May 2026.

Builds on

25 of 28 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Time Series Forecasting via Reasoning: A Slow-Thinking Approach with Reinforcement Fine-Tuned LLMs12 Jun
  2. 2024ChatTS: Aligning Time Series with LLMs via Synthetic Data for Enhanced Understanding and Reasoning4 Dec
  3. 2024A Picture is Worth A Thousand Numbers: Enabling LLMs Reason about Time Series via Visualization9 Nov
  4. 2024Language Models Still Struggle to Zero-shot Reason about Time Series17 Apr

Other references · 21

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  3. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  4. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  5. 2022Scaling Instruction-Finetuned Language Modelscited by 2,997
  6. 2022STaR: Bootstrapping Reasoning With Reasoningcited by 895
  7. 2023A decoder-only foundation model for time-series forecastingcited by 650
  8. 2024Unified Training of Universal Time Series Forecasting Transformerscited by 588
  9. 2025Sundial: A Family of Highly Capable Time Series Foundation Modelscited by 168
  10. 2024Are Language Models Actually Useful for Time Series Forecasting?cited by 163
  11. 2025Chronos-2: From Univariate to Universal Forecastingcited by 158
  12. 2024Context is Key: A Benchmark for Forecasting with Essential Textual Informationcited by 78
9 more
  1. 2025Time-MQA: Time Series Multi-Task Question Answering with Context Enhancementcited by 74
  2. 2024Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysiscited by 65
  3. 2025Language in the Flow of Time: Time-Series-Paired Texts Weaved into a Unified Temporal Narrativecited by 35
  4. 2025Aurora: Towards Universal Generative Multimodal Time Series Forecastingcited by 28
  5. 2024Time-Series Forecasting for Out-of-Distribution Generalization Using Invariant Learningcited by 26
  6. 2025TimeOmni-1: Incentivizing Complex Reasoning with Time Series in Large Language Modelscited by 24
  7. 2023Large Pre-trained time series models for cross-domain Time series analysis taskscited by 22
  8. 2025How Can Time Series Analysis Benefit From Multiple Modalities? A Survey and Outlookcited by 20
  9. 2026FutureX-Pro: Extending Future Prediction to High-Value Vertical Domainscited by 8

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026JudgeCast: Time Series Forecasting with Experience-Informed Covariate Judgementscited by 0
  2. 2026LineupRL: Verifiable Reinforcement Learning for Time Series Captioning via Caption-to-Series Identificationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025UniDiff: A Unified Diffusion Framework for Multimodal Time Series Forecasting
  2. 2026SCENARIODIFF: A Scenario-level Guidance Framework for Multimodal Time Series Forecasting--Extended Version
  3. 2026TiMi: Empower Time Series Transformers with Multimodal Mixture of Experts
  4. 2026Rethinking the Role of LLMs in Time Series Forecasting
  5. 2023LLM4TS: Aligning Pre-Trained LLMs as Data-Efficient Time-Series Forecasters
  6. 2025BALM-TSF: Balanced Multimodal Alignment for LLM-Based Time Series Forecasting
  7. 2025LLM-Integrated Bayesian State Space Models for Multimodal Time-Series Forecasting
  8. 2025Teaching Time Series to See and Speak: Forecasting with Aligned Visual and Textual Perspectives