Contamination Map Methodology

Time series / Model

Timer-XL (timer-base-84m)

Organisation
Tsinghua University (THUML)
Family
Timer · version Timer-XL (ICLR 2025); HF release timer-base-84m
Released
2024-10-07 (arXiv v1 submission date, https://arxiv.org/abs/2410.04803 (HF repo thuml/timer-base-84m created 2024-11-23))
Checkpoints
thuml/timer-base-84m
Weights
Open · Apache-2.0
Links
PaperModel cardCode

Training data disclosure: Partial list

Level determined for 57% of benchmark cells.

“In this section, we conduct univariate pre-training on UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024) and evaluate zero-shot performance on benchmarks from Wu et al. (2022).”
arxiv.org · paper · Section 4.x Large-scale pre-training, Setups · retrieved 2026-10-15
“This version is univariate pre-trained on **260B** time points with **84M** parameters”
huggingface.co · model card · Model card intro · retrieved 2026-10-15
“The model is mostly built from the Internet public time series dataset”
huggingface.co · model card · Acknowledgments · retrieved 2026-10-15

Training corpora

Declared exclusions

Benchmark cells

GIFT-Eval

Unknown Jena Weather (GIFT-Eval) · training data not disclosed in enough detail

Unknown BizITObs - Application · training data not disclosed in enough detail

Unknown BizITObs - Service · training data not disclosed in enough detail

Unknown BizITObs - L2C · training data not disclosed in enough detail

Unknown Bitbrains - Fast Storage · training data not disclosed in enough detail

Unknown Bitbrains - rnd · training data not disclosed in enough detail

Seen, split unclear Restaurant (Recruit Restaurant Visitor Forecasting) · the dataset was used; the source does not say which part (via LOTSA (Large-scale Open Time Series Archive))

Evidence (3)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Restaurant”
arxiv.org · paper · Appendix A, Table 17 (Others) · retrieved 2026-10-15

Unknown ETT1 (GIFT-Eval) · training data not disclosed in enough detail

Unknown ETT2 (GIFT-Eval) · training data not disclosed in enough detail

Seen, split unclear Loop Seattle (LibCity) · the dataset was used; the source does not say which part (via LOTSA (Large-scale Open Time Series Archive))

Evidence (3)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Loop Seattle”
arxiv.org · paper · Appendix A, Table 13 (LibCity) · retrieved 2026-10-15

Seen, split unclear SZ-Taxi (LibCity) · the dataset was used; the source does not say which part (via LOTSA (Large-scale Open Time Series Archive))

Evidence (3)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“SZ-Taxi”
arxiv.org · paper · Appendix A, Table 13 (LibCity) · retrieved 2026-10-15

Seen, split unclear M_DENSE (LibCity) · the dataset was used; the source does not say which part (via LOTSA (Large-scale Open Time Series Archive))

Evidence (3)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Rotterdam”
arxiv.org · paper · Appendix A, Table 13 (LibCity) · retrieved 2026-10-15

Unknown Solar (LSTNet, 137 Alabama PV plants) · training data not disclosed in enough detail

Seen, split unclear Hierarchical Sales (Mancuso et al., 2020) · the dataset was used; the source does not say which part (via LOTSA (Large-scale Open Time Series Archive))

Evidence (3)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Hierarchical Sales”
arxiv.org · paper · Appendix A, Table 17 (Others) · retrieved 2026-10-15

Test data seen M4 Yearly · the whole dataset was used (via LOTSA (Large-scale Open Time Series Archive))

Evidence (5)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“M4 Yearly”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“For the following datasets, we include their entirety in LOTSA without a held-out test set for the following reasons:”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15
“M1 Yearly, M1 Quarterly, M3 Yearly, M3 Quarterly, M4 Yearly, M4 Quarterly, Tourism Yearly: Some time series in these datasets are too short after train/test splits, thus we do not split them”
arxiv.org · paper · Appendix A, Monash exceptions list · retrieved 2026-10-15

Test data seen M4 Quarterly · the whole dataset was used (via LOTSA (Large-scale Open Time Series Archive))

Evidence (5)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“M4 Quarterly”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“For the following datasets, we include their entirety in LOTSA without a held-out test set for the following reasons:”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15
“M1 Yearly, M1 Quarterly, M3 Yearly, M3 Quarterly, M4 Yearly, M4 Quarterly, Tourism Yearly: Some time series in these datasets are too short after train/test splits, thus we do not split them”
arxiv.org · paper · Appendix A, Monash exceptions list · retrieved 2026-10-15

Seen, split unclear M4 Monthly · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“M4 Monthly”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Seen, split unclear M4 Weekly · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“M4 Weekly”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Seen, split unclear M4 Daily · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“M4 Daily”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Seen, split unclear M4 Hourly · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“M4 Hourly”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Seen, split unclear Hospital · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Hospital”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Seen, split unclear COVID Deaths · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“COVID Deaths”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Seen, split unclear US Births · the dataset was used; the source does not say which part (via Unified Time Series Dataset (UTSD))

Evidence (3)
“We adopt the largest volume that encompasses 1 billion time points for pre-training.”
arxiv.org · paper · Appendix B.1, dataset (10) UTSD · retrieved 2026-10-15
“thuml/UTSD”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“US Births | 0.00M”
huggingface.co · dataset card · Dataset Description table · retrieved 2026-10-15

Seen, split unclear Saugeen River Flow · the dataset was used; the source does not say which part (via Unified Time Series Dataset (UTSD))

Evidence (3)
“We adopt the largest volume that encompasses 1 billion time points for pre-training.”
arxiv.org · paper · Appendix B.1, dataset (10) UTSD · retrieved 2026-10-15
“thuml/UTSD”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Saugen River Flow | 0.02M”
huggingface.co · dataset card · Dataset Description table · retrieved 2026-10-15

Seen, split unclear Temperature Rain · the dataset was used; the source does not say which part (via Unified Time Series Dataset (UTSD))

Evidence (3)
“We adopt the largest volume that encompasses 1 billion time points for pre-training.”
arxiv.org · paper · Appendix B.1, dataset (10) UTSD · retrieved 2026-10-15
“thuml/UTSD”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Temperature Rain | 23.25M”
huggingface.co · dataset card · Dataset Description table · retrieved 2026-10-15

Seen, split unclear KDD Cup 2018 · the dataset was used; the source does not say which part (via Unified Time Series Dataset (UTSD))

Evidence (3)
“We adopt the largest volume that encompasses 1 billion time points for pre-training.”
arxiv.org · paper · Appendix B.1, dataset (10) UTSD · retrieved 2026-10-15
“thuml/UTSD”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“KDD Cup 2018 | 2.94M”
huggingface.co · dataset card · Dataset Description table · retrieved 2026-10-15

Seen, split unclear Car Parts · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Car Parts”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Unknown ElectricityLoadDiagrams20112014 (UCI) · training data not disclosed in enough detail

Long-horizon forecasting (ETT, Electricity, Traffic, Weather, Exchange, ILI)

Not seen ETTh1 · the authors state the dataset was excluded from training

Evidence (2)
“We conduct zero-shot evaluations on datasets that are not included during the pre-training of corresponding models.”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Table 13 provides the full results of zero-shot forecasting on the benchmark from Wu et al. (2022).”
arxiv.org · paper · Appendix E.2 / Table 13 · retrieved 2026-10-15

Not seen ETTh2 · the authors state the dataset was excluded from training

Evidence (2)
“We conduct zero-shot evaluations on datasets that are not included during the pre-training of corresponding models.”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Table 13 provides the full results of zero-shot forecasting on the benchmark from Wu et al. (2022).”
arxiv.org · paper · Appendix E.2 / Table 13 · retrieved 2026-10-15

Not seen ETTm1 · the authors state the dataset was excluded from training

Evidence (2)
“We conduct zero-shot evaluations on datasets that are not included during the pre-training of corresponding models.”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Table 13 provides the full results of zero-shot forecasting on the benchmark from Wu et al. (2022).”
arxiv.org · paper · Appendix E.2 / Table 13 · retrieved 2026-10-15

Not seen ETTm2 · the authors state the dataset was excluded from training

Evidence (2)
“We conduct zero-shot evaluations on datasets that are not included during the pre-training of corresponding models.”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Table 13 provides the full results of zero-shot forecasting on the benchmark from Wu et al. (2022).”
arxiv.org · paper · Appendix E.2 / Table 13 · retrieved 2026-10-15

Not seen Electricity (ECL, 321 clients) · the authors state the dataset was excluded from training

Evidence (2)
“We conduct zero-shot evaluations on datasets that are not included during the pre-training of corresponding models.”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Table 13 provides the full results of zero-shot forecasting on the benchmark from Wu et al. (2022).”
arxiv.org · paper · Appendix E.2 / Table 13 · retrieved 2026-10-15

Seen, split unclear Traffic (PeMS San Francisco Bay, 862 sensors) · a training split was used; its overlap with this benchmark's test data is not established (via LOTSA (Large-scale Open Time Series Archive))

Evidence (4)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“Traffic Hourly”
arxiv.org · paper · Appendix A, Table 14 (Monash) · retrieved 2026-10-15
“apart from the exceptions listed below, we only include the train region in LOTSA, by holding out the final forecast horizon as the test set.”
arxiv.org · paper · Appendix A, Monash paragraph · retrieved 2026-10-15

Not seen Weather (Jena, 2020; Autoformer) · the authors state the dataset was excluded from training

Evidence (2)
“We conduct zero-shot evaluations on datasets that are not included during the pre-training of corresponding models.”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Table 13 provides the full results of zero-shot forecasting on the benchmark from Wu et al. (2022).”
arxiv.org · paper · Appendix E.2 / Table 13 · retrieved 2026-10-15

Unknown Exchange-Rate · training data not disclosed in enough detail

Related data seen ILI (CDC influenza-like illness, national) · related dataset CDC FluView ILINet (LOTSA) was used (via LOTSA (Large-scale Open Time Series Archive))

Evidence (6)
“datasets of UTSD (Liu et al., 2024c) and LOTSA (Woo et al., 2024). The patch size of Timer-XL is set as 96”
arxiv.org · paper · Appendix E.2 · retrieved 2026-10-15
“Salesforce/lotsa_data”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“CDC Fluview ILINet”
arxiv.org · paper · Appendix A, Table 17 (Others) · retrieved 2026-10-15
“includes the weekly recorded influenza-like illness (ILI) patients data from Centers for Disease Control and Prevention of the United States between 2002 and 2021”
arxiv.org · paper · Section 4 (Datasets) · retrieved 2026-10-15
“https://gis.cdc.gov/grasp/fluview/fluportaldashboard.html”
arxiv.org · paper · Section 4, footnote 4 · retrieved 2026-10-15
“CDC Fluview ILINet CDC Fluview WHO NREVSS”
arxiv.org · paper · Table 14 (dataset column) · retrieved 2026-10-15

Time-series anomaly detection

Unknown TSB-AD-U (TSB-AD univariate) · training data not disclosed in enough detail

Unknown TSB-AD-M (TSB-AD multivariate) · training data not disclosed in enough detail

Unknown UCR Time Series Anomaly Archive (Hexagon ML/UCR, KDD Cup 2021) · training data not disclosed in enough detail

Unknown Server Machine Dataset (SMD) · training data not disclosed in enough detail

Unknown Mars Science Laboratory (MSL) telemetry anomaly dataset · training data not disclosed in enough detail

Unknown Soil Moisture Active Passive (SMAP) telemetry anomaly dataset · training data not disclosed in enough detail

Unknown Pooled Server Metrics (PSM) · training data not disclosed in enough detail

Unknown Secure Water Treatment (SWaT) · training data not disclosed in enough detail

Unknown Yahoo S5 Labeled Anomaly Detection Dataset (Yahoo Webscope S5) · training data not disclosed in enough detail

Unknown Numenta Anomaly Benchmark (NAB) · training data not disclosed in enough detail

Time-series classification

Seen, split unclear UCR Time Series Classification Archive (2018) · the dataset was used; the source does not say which part (via Unified Time Series Dataset (UTSD))

Evidence (6)
“We adopt the largest volume that encompasses 1 billion time points for pre-training.”
arxiv.org · paper · Appendix B.1, dataset (10) UTSD · retrieved 2026-10-15
“thuml/UTSD”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“PigArtPressure | 0.62M”
huggingface.co · dataset card · Dataset Description table · retrieved 2026-10-15
“PigArtPressure | 0.62M”
huggingface.co · dataset card · Dataset Description table (source column [3]) · retrieved 2026-10-15
“[3]: [The UCR Time Series Archive](https://arxiv.org/abs/1810.07758)”
huggingface.co · dataset card · Dataset Description references · retrieved 2026-10-15
“Each series representing a variate will be divided into training and validation splits at a ratio of 9:1 for pre-training.”
arxiv.org · paper · Section 3 (pre-training strategy) · retrieved 2026-10-15

Seen, split unclear UEA Multivariate Time Series Classification Archive (2018) · the dataset was used; the source does not say which part (via Unified Time Series Dataset (UTSD))

Evidence (6)
“We adopt the largest volume that encompasses 1 billion time points for pre-training.”
arxiv.org · paper · Appendix B.1, dataset (10) UTSD · retrieved 2026-10-15
“thuml/UTSD”
huggingface.co · model card · YAML metadata `datasets` · retrieved 2026-10-15
“MotorImagery | 72.58M”
huggingface.co · dataset card · Dataset Description table · retrieved 2026-10-15
“MotorImagery | 72.58M”
huggingface.co · dataset card · Dataset Description table (source column [3]) · retrieved 2026-10-15
“[3]: [The UCR Time Series Archive](https://arxiv.org/abs/1810.07758)”
huggingface.co · dataset card · Dataset Description references · retrieved 2026-10-15
“Each series representing a variate will be divided into training and validation splits at a ratio of 9:1 for pre-training.”
arxiv.org · paper · Section 3 (pre-training strategy) · retrieved 2026-10-15

Notes

Model card: 'introduced in [Timer] and enhanced by [Timer-XL]'; Timer-XL README says the HF model is its univariate pre-trained release. Disclosure set to partial_list because the card's 260B time points exceeds UTSD-12G (~0.87B train points) + LOTSA (~231B) listed in Timer-XL Table 9, and the card says 'mostly' public data. Timer-XL also pre-trains a separate domain-specific model on its own ERA5-Large (4920 stations), not the released checkpoint. In Timer-XL Table 13 Time-MoE's ECL cell is a dash with footnote 'Dataset for pre-training is not evaluated on corresponding models' -- third-party confirmation that Time-MoE saw ECL.

Sources