---
license: apache-2.0
library_name: onnx
tags:
  - time-series-forecasting
  - chronos
  - onnx
  - openstef
---

# chronos-2 — ONNX checkpoints

ONNX export of [`amazon/chronos-2`](https://huggingface.co/amazon/chronos-2) for
[OpenSTEF](https://github.com/OpenSTEF/openstef), produced by
[`openstef-checkpoints`](https://github.com/OpenSTEF/openstef-checkpoints).

OpenSTEF resolves these via `HubCheckpoint`; the metadata file next to each weights file drives the
inference path. Each variant was checked against the torch reference on representative inputs before
publishing.

## Variants

| File | Precision | Static shapes | Max deviation vs torch |
|---|---|---|---|
| `chronos-2_static.onnx` | fp32 | yes | 1.788e-07 |
| `chronos-2.onnx` | fp32 | no | 2.98e-08 |
| `chronos-2_int8.onnx` | int8 | no | 9.677e-05 |


Pick by deployment target: **static fp32** is the portable, CoreML-eligible default; **int8** for size;
**dynamic** when context/horizon must vary.

## Provenance

- Source model: `amazon/chronos-2` @ `unknown`
- Exporter: `openstef-checkpoints` @ `f1ea4ab27067a7ce8b8ec79d6dd0ae9656428e11`
- Tooling: onnx=1.22.0 onnxruntime=1.27.0 torch=2.12.1+cpu
- Exported: 2026-06-19T11:18:21+00:00

## License and attribution

These ONNX checkpoints are derived from
[`amazon/chronos-2`](https://huggingface.co/amazon/chronos-2) and released under the same
license, **apache-2.0**. Attribution and all rights to the model weights remain with the
upstream authors; if you use the model in research, please cite their work.

**Modifications.** The original weights were converted from PyTorch to ONNX and exported in several
shape and precision variants (static or dynamic shapes; fp32, fp16, or int8). The weights were not
retrained, fine-tuned, or otherwise altered.

The tooling that produced these files
([`openstef-checkpoints`](https://github.com/OpenSTEF/openstef-checkpoints)) is licensed MPL-2.0,
which does not extend to the weights.