Version 1.2.0

We are happy to announce the Uni2TS 1.2.0 release. In this version, we provide the code for Moirai-MoE which is our latest model. The model weights can be downloaded from huggingface: small and base.

Besides Moirai-MoE, we also provide more evaluation files to facilite the evaluation for some foundation models (e.g., Chronos, TimesFM, VisionTS). Meanwhile, of course, some enhancements and fixes (e.g., save huggingface checkpoint, training resume) are contained in this version :)

What's Changed in the Codebase

Moirai-MoE

update for moirai-moe by @liuxu77 in #139
Update readme file by @liuxu77 in #140
update readme and notebooks for moe by @liuxu77 in #142

More Evaluation files

Evaluate chronos on monash by @liu-jc in #72
update evaluation code and scripts by @liu-jc in #113
Scripts and code to evaluate timesfm by @liu-jc in #119
add VisionTS scripts and code by @Keytoyze in #120
fix VisionTS context len issue by @Keytoyze in #124

Enhancements and Fixes

Add ckpt_path config to resume training by @gorold in #67
Documentation by @gorold in #71
Change to python3 for black in pre-commit by @liu-jc in #73
add config file for lightning checkpoint by @gorold in #96
change the version name and the project name for pypi by @liu-jc in #106
Implement a new callback HugginfaceCheckpoint by @liu-jc in #109
Manually set 'freq' for specific datasets by @zqiao11 in #110
Update README and config files with Moirai-1.1-R and fix Moirai-1.0-R model weights by @gorold in #111
update ipynb examples with 1.1 models by @liu-jc in #112
update readme by @liu-jc in #131
update README by @liu-jc in #152
update version by @liu-jc in #153
Update init.py adding version by @liu-jc in #155
update workflows by @liu-jc in #157, #158