v1.2.1

google-research/timesfmv1.2.1Oct 18, 2024by siriuz42

AI Summary

This release adds PyTorch support for TimesFM inference, enabling users to choose between JAX and PyTorch backends. It introduces separate dependency installations allowing users to install either the Pax/JAX version or the PyTorch version of TimesFM.

Key Highlights

PyTorch support for TimesFM inference
Separate dependency installation: timesfm[pax] for JAX, timesfm[torch] for PyTorch
PyTorch checkpoint for 200m model available on Hugging Face
Modular package structure for different backend preferences

New Features

PyTorch inference support
Separate pax and torch package variants
timesfm-1.0-200m-pytorch checkpoint

Full Release Notes

Changes:
----
- PyTorch support for TimesFM inference.

PyPI Release:
----

v1.2.1: Support separate dependencies for `pax` and `torch` versions of TimesFM:
  - `pip install timesfm[pax]` for the `pax` version and `jax` checkpoints.
  - `pip install timesfm[torch]` for the `torch` version and checkpoints.
  - See the updated [README](https://github.com/google-research/timesfm?tab=readme-ov-file#usage) for the usage.

Checkpoints:
----
The PyTorch checkpoint for the 200m model is available on Hugging Face:
- https://huggingface.co/google/timesfm-1.0-200m-pytorch

Full Changelog:
----
https://github.com/google-research/timesfm/commits/v1.2.1