Contamination MapTime series / Model
Lag-Llama
- Organisation
- Lag-Llama authors (ServiceNow Research, Mila, Morgan Stanley et al.)
- Family
- Lag-Llama
- Released
- 2023-10-12 (arXiv v1 submission date (https://arxiv.org/abs/2310.08278); open weights released 2024-02-07 per the model card ('7-Feb-2024: We released Lag-Llama, with open-source model checkpoints').)
- Checkpoints
- time-series-foundation-models/Lag-Llama
- Weights
- Open · apache-2.0
- Links
- PaperModel cardCode
Training data disclosure: Full list
Level determined for 100% of benchmark cells.
- TestTest data seen0
- SeenSeen, split unclear7
- TrainTraining data seen0
- Related data seen4
- NoNot seen38
- ?Unknown0
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Other declared training data
-
Traffic (PeMS San Francisco Bay, 862 sensors) · split: train
Appendix A: 862 hourly series of SF Bay Area freeway occupancy 2015-2016 (Godahewa et al., 2021), i.e. the same source data as the long-horizon Traffic dataset. Train split as defined in the (Monash/GluonTS) dataset, not the Informer split.
- Uber TLC Hourly · split: train
- Australian Electricity Demand · split: train
-
Electricity (ECL, 321 clients) · split: train
Appendix A: 321 clients, hourly, 2012-2014 (Godahewa et al., 2021), i.e. the same source data as the long-horizon Electricity dataset. Train split as defined in the (Monash/GluonTS) dataset, not necessarily the Informer split.
- London Smart Meters · split: train
-
Solar (LSTNet, 137 Alabama PV plants) · split: train
Appendix A: 137 series of 10-minute solar power production in Alabama during 2006 (Monash solar_10_minutes). Appendix A: 137 Alabama plants, 10-minute (Monash solar_10_minutes).
- Wind Farms · split: train
-
ETTh1 · split: train
Train split 'as defined in the dataset'; the paper does not say this is the Informer/long-horizon split.
-
ETTh2 · split: train
Train split 'as defined in the dataset'; the paper does not say this is the Informer/long-horizon split.
-
ETTm1 · split: train
Train split 'as defined in the dataset'; the paper does not say this is the Informer/long-horizon split.
- KDD Cup 2018 · split: train
- Sunspot · split: train
- Beijing Multisite · split: train
- UCI · split: train
- CPU Limit Minute · split: train
- CPU Usage Minute · split: train
- Function Delay Minute · split: train
- Instances Minute · split: train
- Memory Limit Minute · split: train
- Memory Usage Minute · split: train
Declared exclusions
- Dataset ETTm2 · entirely excluded
- Dataset Exchange-Rate · entirely excluded
Benchmark cells
GIFT-Eval
Not seen Jena Weather (GIFT-Eval) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen BizITObs - Application · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen BizITObs - Service · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen BizITObs - L2C · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Bitbrains - Fast Storage · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Bitbrains - rnd · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Restaurant (Recruit Restaurant Visitor Forecasting) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
ETT1 (GIFT-Eval) · related dataset ETTh1 was used
Evidence (4)
“ETT H1 ETT H2”
“When pretraining, we train on a combined dataset formed out of the train split of each dataset”
“From Informer (Zhou et al., 2021) we utilize ETT1 and ETT2 datasets, which denote electricity transformer temperature”
“ETTh1&ETTh2 7 17,420”
ETT2 (GIFT-Eval) · related dataset ETTh2 was used
Evidence (4)
“ETT H2 ETT M1”
“When pretraining, we train on a combined dataset formed out of the train split of each dataset”
“From Informer (Zhou et al., 2021) we utilize ETT1 and ETT2 datasets, which denote electricity transformer temperature”
“ETTh1&ETTh2 7 17,420”
Not seen Loop Seattle (LibCity) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen SZ-Taxi (LibCity) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen M_DENSE (LibCity) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Seen, split unclear Solar (LSTNet, 137 Alabama PV plants) · a training split was used; its overlap with this benchmark's test data is not established
Not seen Hierarchical Sales (Mancuso et al., 2020) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen M4 Yearly · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen M4 Quarterly · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen M4 Monthly · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen M4 Weekly · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen M4 Daily · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen M4 Hourly · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Hospital · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen COVID Deaths · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen US Births · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Saugeen River Flow · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Temperature Rain · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Seen, split unclear KDD Cup 2018 · a training split was used; its overlap with this benchmark's test data is not established
Not seen Car Parts · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
ElectricityLoadDiagrams20112014 (UCI) · related dataset Electricity (ECL, 321 clients) was used
Evidence (5)
“Electricity Hourly”
“When pretraining, we train on a combined dataset formed out of the train split of each dataset”
“The raw dataset is in https://archive.ics.uci.edu/ml/datasets/ElectricityLoadDiagrams20112014.”
“Because the some dimensions are equal to 0. So we eliminate the records in 2011. Final we get data contains electircity consumption of 321 clients from 2012 to 2014.”
“https://archive.ics.uci.edu/ml/datasets/ElectricityLoadDiagrams20112014”
Long-horizon forecasting (ETT, Electricity, Traffic, Weather, Exchange, ILI)
Seen, split unclear ETTh1 · a training split was used; its overlap with this benchmark's test data is not established
Seen, split unclear ETTh2 · a training split was used; its overlap with this benchmark's test data is not established
Seen, split unclear ETTm1 · a training split was used; its overlap with this benchmark's test data is not established
ETTm2 · related dataset ETTh2 was used; the authors state this dataset was excluded
Evidence (5)
“ETT H2 ETT M1”
“When pretraining, we train on a combined dataset formed out of the train split of each dataset”
“we also provide the hourly-level variants for fast development (marked by *h*), i.e. ETT-small-h1 and ETT-small-h2.”
“ETT M2 Banking & Econ”
“Table 3: Datasets used in the pretraining corpus and the unseen datasets on which we evaluate”
Seen, split unclear Electricity (ECL, 321 clients) · a training split was used; its overlap with this benchmark's test data is not established
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
Not seen Weather (Jena, 2020; Autoformer) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Exchange-Rate · the authors state the dataset was excluded from training
Not seen ILI (CDC influenza-like illness, national) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Time-series anomaly detection
Not seen TSB-AD-U (TSB-AD univariate) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen TSB-AD-M (TSB-AD multivariate) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen UCR Time Series Anomaly Archive (Hexagon ML/UCR, KDD Cup 2021) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Server Machine Dataset (SMD) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Mars Science Laboratory (MSL) telemetry anomaly dataset · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Soil Moisture Active Passive (SMAP) telemetry anomaly dataset · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Pooled Server Metrics (PSM) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Secure Water Treatment (SWaT) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Yahoo S5 Labeled Anomaly Detection Dataset (Yahoo Webscope S5) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen Numenta Anomaly Benchmark (NAB) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Time-series classification
Not seen UCR Time Series Classification Archive (2018) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Not seen UEA Multivariate Time Series Classification Archive (2018) · absent from the complete list of training data
Evidence (2)
“We collate a diverse corpus of 27 time series datasets from several sources across six different semantically grouped domains”
“Table 3 further presents if a dataset was present in the pretraining or downstream testing corpora in our work.”
Notes
Pretraining used only the train split of each of the 20 pretraining datasets (Appendix B); the split is 'as defined in the dataset' (GluonTS/Monash-style), not stated to be the Informer long-horizon split. ETTh1/ETTh2/ETTm1 and the Monash versions of Electricity (321 clients) and Traffic (862 series) were in pretraining; ETTm2, Weather (Monash, Australian), Exchange Rate, Pedestrian Counts, Beijing PM2.5 and two Huawei cloud series were held out as unseen. No synthetic data. No newer Lag-Llama release found as of 2026-10-15. Other stated exclusions, not tracked here: Pedestrian Counts (all); Weather (all); Beijing PM2.5 (all); Requests Minute (all); Platform Delay Minute (all).
Sources
- Lag-Llama: Towards Foundation Models for Probabilistic Time Series Forecasting (arXiv v3) · paper · retrieved 2026-10-15
- Lag-Llama model card · model card · retrieved 2026-10-15