Saved in:
Bibliographic Details
Main Authors: Zhang, Xinyu, Feng, Shanshan, Li, Xutao, Lin, Kenghong, Li, Fan, Jia, Pengfei
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.08818
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912946593988608
author Zhang, Xinyu
Feng, Shanshan
Li, Xutao
Lin, Kenghong
Li, Fan
Jia, Pengfei
author_facet Zhang, Xinyu
Feng, Shanshan
Li, Xutao
Lin, Kenghong
Li, Fan
Jia, Pengfei
contents Using pre-trained large language models (LLMs) as a backbone for time series prediction has recently attracted growing research interest. Existing approaches typically split time series into patches, map them to the token space of LLMs via a Tokenizer, process the tokens through a frozen or fine-tuned LLM backbone, and then reconstruct numerical forecasts using a Detokenizer. However, the actual effectiveness of LLMs for time series forecasting remains under debate. We observe that when trained and evaluated on small datasets, these Tokenizer-Detokenizer pairs often overfit to the specific data distribution, thereby masking the intrinsic predictive capability of the LLM backbone. To investigate the inherent potential of LLMs in this context, we design three models with identical architectures but distinct pre-training strategies. By leveraging large-scale pre-training, we obtain more unbiased Tokenizer-Detokenizer pairs that are seamlessly integrated with the LLM backbone. Through controlled experiments, we evaluate the zero-shot and few-shot forecasting performance of the LLM, offering insights into its true capabilities. Our extensive experiments reveal that, although the LLM backbone shows some promise, its performance remains limited and does not consistently surpass that of models specifically trained on large-scale time series data. Our source code is publicly available in the repository: https://github.com/SiriZhang45/LLM4TS.
format Preprint
id arxiv_https___arxiv_org_abs_2504_08818
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Tokenizer Bias to Backbone Capability: A Controlled Study of LLMs for Time Series Forecasting
Zhang, Xinyu
Feng, Shanshan
Li, Xutao
Lin, Kenghong
Li, Fan
Jia, Pengfei
Machine Learning
Artificial Intelligence
Using pre-trained large language models (LLMs) as a backbone for time series prediction has recently attracted growing research interest. Existing approaches typically split time series into patches, map them to the token space of LLMs via a Tokenizer, process the tokens through a frozen or fine-tuned LLM backbone, and then reconstruct numerical forecasts using a Detokenizer. However, the actual effectiveness of LLMs for time series forecasting remains under debate. We observe that when trained and evaluated on small datasets, these Tokenizer-Detokenizer pairs often overfit to the specific data distribution, thereby masking the intrinsic predictive capability of the LLM backbone. To investigate the inherent potential of LLMs in this context, we design three models with identical architectures but distinct pre-training strategies. By leveraging large-scale pre-training, we obtain more unbiased Tokenizer-Detokenizer pairs that are seamlessly integrated with the LLM backbone. Through controlled experiments, we evaluate the zero-shot and few-shot forecasting performance of the LLM, offering insights into its true capabilities. Our extensive experiments reveal that, although the LLM backbone shows some promise, its performance remains limited and does not consistently surpass that of models specifically trained on large-scale time series data. Our source code is publicly available in the repository: https://github.com/SiriZhang45/LLM4TS.
title From Tokenizer Bias to Backbone Capability: A Controlled Study of LLMs for Time Series Forecasting
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2504.08818