ViTime
Official pretrained checkpoint for ViTime: Foundation Model for Time Series Forecasting Powered by Vision Intelligence.
ViTime converts numerical time series into binary images and performs point and probabilistic forecasting with a vision-based architecture.
Checkpoint
| File | Size | SHA-256 |
|---|---|---|
ViTime_Model.pth |
296,778,986 bytes | 6513b03b352163337c333526fd3634b07db789665cd9f87648f9661efe0cff1a |
The stable release revision is v1.0.0.
Download
Public downloads do not require a Hugging Face account or access token.
from huggingface_hub import hf_hub_download
checkpoint_path = hf_hub_download(
repo_id="IkeYEUNG/ViTime",
filename="ViTime_Model.pth",
revision="v1.0.0",
)
print(checkpoint_path)
Command-line download:
hf download IkeYEUNG/ViTime ViTime_Model.pth --revision v1.0.0
Use with the official GitHub repository
git clone https://github.com/IkeYang/ViTime.git
cd ViTime
python -m pip install -r requirements.txt
The official code downloads this v1.0.0 checkpoint automatically on first
use and reuses the Hugging Face cache on later runs:
import numpy as np
from main import ViTimePrediction
x = np.sin(np.arange(512) / 10)
model = ViTimePrediction(device="cuda:0", model_name="MAE", lookbackRatio=None)
prediction = model.prediction(x, future_length=720)
print(prediction.shape)
See the GitHub repository for installation requirements and complete point/probabilistic forecasting examples.
Checkpoint security
This release preserves the original PyTorch .pth checkpoint format for compatibility with the official code. Load serialized PyTorch checkpoints only from trusted sources and pin the v1.0.0 revision for reproducible use.
Citation
@article{yang2025vitime,
title={{ViTime}: Foundation Model for Time Series Forecasting Powered by Vision Intelligence},
author={Yang, Luoxiao and Wang, Yun and Fan, Xinqi and Cohen, Israel and Chen, Jingdong and Zhang, Zijun},
journal={Transactions on Machine Learning Research},
year={2025},
url={https://openreview.net/forum?id=XInsJDBIkp},
note={Published in Transactions on Machine Learning Research (10/2025)}
}