Time Series Forecasting on ETTm2 (5% train)
0.169MSETime-VLM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Time-VLMforecast_horizon=96, training_data_ratio=5%2026.05 | 0.169 | 0.26 | |
| Time-LLMforecast_horizon=96, training_data_ratio=5%2026.05 | 0.174 | 0.261 | |
| VisionTSforecast_horizon=96, training_data_ratio=5%2026.05 | 0.196 | 0.277 | |
| GPT4TSforecast_horizon=96, training_data_ratio=5%2026.05 | 0.199 | 0.28 | |
| PatchTSTforecast_horizon=96, training_data_ratio=5%2026.05 | 0.206 | 0.288 | |
| SSDAforecast_horizon=96, training_data_ratio=5%2026.05 | 0.207 | 0.272 | |
| Time-LLMforecast_horizon=192, training_data_ratio=5%2026.05 | 0.215 | 0.287 | |
| TimesNetforecast_horizon=96, training_data_ratio=5%2026.05 | 0.22 | 0.299 | |
| Time-VLMforecast_horizon=192, training_data_ratio=5%2026.05 | 0.224 | 0.298 | |
| FEDformerforecast_horizon=96, training_data_ratio=5%2026.05 | 0.229 | 0.32 | |
| DLinearforecast_horizon=96, training_data_ratio=5%2026.05 | 0.236 | 0.326 | |
| TTM + CoRAModel Category=Pre-trained, Plugin=CoRA2026.03 | 0.254 | — | |
| VisionTSforecast_horizon=192, training_data_ratio=5%2026.05 | 0.255 | 0.316 | |
| GPT4TSforecast_horizon=192, training_data_ratio=5%2026.05 | 0.256 | 0.316 | |
| FiCoTSInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=LLM-based2025.11 | 0.257 | 0.317 | |
| Timer + CoRAModel Category=Pre-trained, Plugin=CoRA2026.03 | 0.257 | — | |
| TTMModel Category=Pre-trained, Plugin=None2026.03 | 0.259 | — | |
| TimeVLMInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=LLM-based2025.11 | 0.262 | 0.323 | |
| TimerModel Category=Pre-trained, Plugin=None2026.03 | 0.262 | — | |
| Time-VLMRatio=5%2026.05 | 0.262 | 0.323 | |
| CALF + CoRAModel Category=LLM-based, Plugin=CoRA2026.03 | 0.264 | — | |
| PatchTSTforecast_horizon=192, training_data_ratio=5%2026.05 | 0.264 | 0.324 | |
| GPT4TS + CoRAModel Category=LLM-based, Plugin=CoRA2026.03 | 0.269 | — | |
| CALFModel Category=LLM-based, Plugin=None2026.03 | 0.272 | — | |
| Time-LLMforecast_horizon=336, training_data_ratio=5%2026.05 | 0.273 | 0.33 | |
| Time-LLMRatio=5%2026.05 | 0.274 | 0.323 | |
| Moment + CoRAModel Category=Pre-trained, Plugin=CoRA2026.03 | 0.275 | — | |
| GPT4TSModel Category=LLM-based, Plugin=None2026.03 | 0.279 | — | |
| MomentModel Category=Pre-trained, Plugin=None2026.03 | 0.279 | — | |
| TIME-LLMParameters (M)=6.46, Memory (MiB)=3907, Eff.*MSE=0.39, Eff.*MAE=0.38, Few-shot (5% data)=true2026.03 | 0.28 | 0.32 | |
| UniTime + CoRAModel Category=LLM-based, Plugin=CoRA2026.03 | 0.281 | — | |
| SSDARatio=5%2026.05 | 0.281 | 0.326 | |
| SSDAforecast_horizon=192, training_data_ratio=5%2026.05 | 0.282 | 0.327 | |
| Time-VLMforecast_horizon=336, training_data_ratio=5%2026.05 | 0.282 | 0.338 | |
| SSDAforecast_horizon=336, training_data_ratio=5%2026.05 | 0.285 | 0.329 | |
| UniTimeModel Category=LLM-based, Plugin=None2026.03 | 0.292 | — | |
| S^2IP-LLMInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=LLM-based2025.11 | 0.296 | 0.342 | |
| One-for-AllParameters (M)=0.55, Memory (MiB)=2.2, Eff.*MSE=3.92, Eff.*MAE=4.17, Few-shot (5% data)=true2026.03 | 0.3 | 0.35 | |
| DLinearforecast_horizon=192, training_data_ratio=5%2026.05 | 0.306 | 0.373 | |
| TimeLLMInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=LLM-based2025.11 | 0.307 | 0.348 | |
| GPT4TSInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=LLM-based2025.11 | 0.308 | 0.346 | |
| VisionTSRatio=5%2026.05 | 0.308 | 0.35 | |
| GPT4TSRatio=5%2026.05 | 0.308 | 0.346 | |
| GPT4TSParameters (M)=11.7, Memory (MiB)=371.0, Eff.*MSE=0.20, Eff.*MAE=0.20, Few-shot (5% data)=true2026.03 | 0.31 | 0.35 | |
| TimesNetforecast_horizon=192, training_data_ratio=5%2026.05 | 0.311 | 0.361 | |
| PatchTSTInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=Transformer-based2025.11 | 0.314 | 0.352 | |
| PatchTSTRatio=5%2026.05 | 0.314 | 0.352 | |
| GPT4TSforecast_horizon=336, training_data_ratio=5%2026.05 | 0.318 | 0.353 | |
| VisionTSforecast_horizon=336, training_data_ratio=5%2026.05 | 0.321 | 0.364 | |
| PatchTSTforecast_horizon=336, training_data_ratio=5%2026.05 | 0.334 | 0.367 | |
| TimesNetforecast_horizon=336, training_data_ratio=5%2026.05 | 0.338 | 0.366 | |
| TStationaryParameters (M)=2.02, Memory (MiB)=13.2, Eff.*MSE=0.84, Eff.*MAE=1.03, Few-shot (5% data)=true2026.03 | 0.34 | 0.37 | |
| TimesNetInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=Non-Transformer-based2025.11 | 0.344 | 0.372 | |
| TimesNetRatio=5%2026.05 | 0.344 | 0.372 | |
| TimesNetParameters (M)=0.63, Memory (MiB)=2.9, Eff.*MSE=2.89, Eff.*MAE=3.39, Few-shot (5% data)=true2026.03 | 0.35 | 0.37 | |
| SSDAforecast_horizon=720, training_data_ratio=5%2026.05 | 0.35 | 0.377 | |
| iTransformerInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=Transformer-based2025.11 | 0.356 | 0.388 | |
| Time-VLMforecast_horizon=720, training_data_ratio=5%2026.05 | 0.375 | 0.397 | |
| FEDformerforecast_horizon=336, training_data_ratio=5%2026.05 | 0.378 | 0.427 | |
| DLinearforecast_horizon=336, training_data_ratio=5%2026.05 | 0.38 | 0.423 | |
| FEDformerInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=Transformer-based2025.11 | 0.381 | 0.404 | |
| FEDformerRatio=5%2026.05 | 0.381 | 0.404 | |
| AutoformerInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=Transformer-based2025.11 | 0.388 | 0.433 | |
| AutoformerParameters (M)=10.53, Memory (MiB)=62.6, Eff.*MSE=0.18, Eff.*MAE=0.20, Few-shot (5% data)=true2026.03 | 0.39 | 0.44 | |
| FEDformerforecast_horizon=192, training_data_ratio=5%2026.05 | 0.394 | 0.361 | |
| DLinearInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=Non-Transformer-based2025.11 | 0.399 | 0.426 | |
| DLinearRatio=5%2026.05 | 0.399 | 0.426 | |
| EFDformerParameters (M)=16.8, Memory (MiB)=87.8, Eff.*MSE=0.12, Eff.*MAE=0.13, Few-shot (5% data)=true2026.03 | 0.4 | 0.41 | |
| Time-LLMforecast_horizon=720, training_data_ratio=5%2026.05 | 0.433 | 0.412 | |
| PatchTSTforecast_horizon=720, training_data_ratio=5%2026.05 | 0.454 | 0.432 | |
| VisionTSforecast_horizon=720, training_data_ratio=5%2026.05 | 0.459 | 0.444 | |
| GPT4TSforecast_horizon=720, training_data_ratio=5%2026.05 | 0.46 | 0.436 | |
| TimesNetforecast_horizon=720, training_data_ratio=5%2026.05 | 0.509 | 0.465 | |
| FEDformerforecast_horizon=720, training_data_ratio=5%2026.05 | 0.523 | 0.51 | |
| ETSformerParameters (M)=5.28, Memory (MiB)=31.4, Eff.*MSE=0.24, Eff.*MAE=0.29, Few-shot (5% data)=true2026.03 | 0.53 | 0.55 | |
| ETSformerInput sequence length=512, Predictive lengths={96, 192, 336, 720}, Training data ratio=5%, Method category=Transformer-based2025.11 | 0.534 | 0.547 | |
| DLinearforecast_horizon=720, training_data_ratio=5%2026.05 | 0.674 | 0.583 | |
| Informerforecast_horizon=336, training_data_ratio=5%2026.05 | 3.561 | 1.473 | |
| Informerforecast_horizon=192, training_data_ratio=5%2026.05 | 3.578 | 1.475 | |
| ReformerParameters (M)=5.80, Memory (MiB)=33.4, Eff.*MSE=0.09, Eff.*MAE=0.17, Few-shot (5% data)=true2026.03 | 3.58 | 1.49 | |
| Informerforecast_horizon=96, training_data_ratio=5%2026.05 | 3.599 | 1.478 | |
| InformerRatio=5%2026.05 | 3.658 | 1.489 | |
| InformerParameters (M)=11.33, Memory (MiB)=65.8, Eff.*MSE=0.04, Eff.*MAE=0.08, Few-shot (5% data)=true2026.03 | 3.66 | 1.49 | |
| Informerforecast_horizon=720, training_data_ratio=5%2026.05 | 3.896 | 1.533 |