Skip to main content

​
Overview

Chronos-Bolt is an optimized version of Chronos designed for fast inference while maintaining high accuracy. It uses knowledge distillation to create a more efficient model suitable for production environments. Repository: amazon/chronos-bolt-small

​
Loading the Model

from samay.model import ChronosBoltModel

repo = "amazon/chronos-bolt-small"
model = ChronosBoltModel(repo=repo)
Chronos-Bolt is available in multiple sizes: chronos-bolt-tiny, chronos-bolt-mini, chronos-bolt-small, chronos-bolt-base.

​
Loading Dataset

from samay.dataset import ChronosBoltDataset

train_dataset = ChronosBoltDataset(
    name="ett",
    datetime_col="date",
    path="./data/ETTh1.csv",
    mode="train",
    batch_size=8
)

val_dataset = ChronosBoltDataset(
    name="ett",
    datetime_col="date",
    path="./data/ETTh1.csv",
    mode="test",
    batch_size=8
)

​
Zero-Shot Forecasting

metrics = model.evaluate(
    val_dataset,
    horizon_len=64,
    quantile_levels=[0.1, 0.5, 0.9],
    metric_only=True
)
print(metrics)

​
Fine-tuning

model.finetune(train_dataset)

# Evaluate the fine-tuned model
metrics = model.evaluate(
    val_dataset,
    horizon_len=64,
    quantile_levels=[0.1, 0.5, 0.9],
    metric_only=True
)

​
Visualization

model.plot(
    val_dataset,
    horizon_len=64,
    quantile_levels=[0.1, 0.5, 0.9]
)

​
Key Differences from Chronos

FeatureChronosChronos-Bolt
SpeedBaseline2-3x faster
Model SizeLargerMore compact
AccuracyHighestVery competitive
Use CaseResearch & high accuracyProduction & speed

​
Performance Characteristics

  • Faster Inference: 2-3x speedup compared to standard Chronos
  • Smaller Memory Footprint: More efficient memory usage
  • Competitive Accuracy: Maintains >95% of full Chronos accuracy
  • Production-Ready: Optimized for deployment

​
When to Use Chronos-Bolt

  • When you need fast inference in production
  • For real-time forecasting applications
  • When computational resources are limited
  • For edge deployment scenarios
  • When you need a good balance of speed and accuracy

​
Model Variants

repo = "amazon/chronos-bolt-tiny"
model = ChronosBoltModel(repo=repo)
# Fastest, smallest model for edge devices

​
Quantile Forecasting

Like Chronos, Chronos-Bolt supports probabilistic forecasting: 
quantile_outputs, mean = model.pipeline.predict_quantiles(
    context=context,
    prediction_length=64,
    quantile_levels=[0.1, 0.5, 0.9]
)

​
Example Notebook

For a complete working example, see:
Chronos-Bolt is ideal for production deployments where inference speed is critical while maintaining high forecasting accuracy.

Build docs developers (and LLMs) love

Get started for freeTalk to us