Views
No views yet
| Attribute | Value |
|---|---|
| Parameters | 710M |
| Architecture | T5 Encoder-Decoder |
| Original Repo | amazon/chronos-t5-large |
| Paper | Chronos: Learning the Language of Time Series |
| Task | Time Series Forecasting |
1import ts_arena
2
3# Load model
4model = ts_arena.load_model("chronos-t5-large")
5
6# Generate forecasts
7import numpy as np
8context = np.random.randn(96) # 96 timesteps of history
9output = model.predict(context, prediction_length=24, num_samples=20)
10
11# Access results
12print(output.predictions.shape) # Point forecasts (median)
13print(output.quantiles[0.5].shape) # Median forecast
14print(output.quantiles[0.1].shape) # 10th percentile
15print(output.quantiles[0.9].shape) # 90th percentile1from chronos import ChronosPipeline
2import torch
3
4pipeline = ChronosPipeline.from_pretrained(
5 "amazon/chronos-t5-large",
6 device_map="cuda",
7 torch_dtype=torch.bfloat16,
8)
9
10context = torch.randn(1, 96) # (batch, time)
11forecast = pipeline.predict(context, prediction_length=24, num_samples=20)| Metric | Value |
|---|---|
| MSE | 4.37 |
| MAE | 1.66 |
| RMSE | 2.09 |
1@article{ansari2024chronos,
2 title={Chronos: Learning the Language of Time Series},
3 author={Ansari, Abdul Fatir and Stella, Lorenzo and Turkmen, Caner and Zhang, Xiyuan and others},
4 journal={arXiv preprint arXiv:2403.07815},
5 year={2024}
6}