timesfm

Run it

$ pip install pfnstudio
$ pfnstudio run timesfm --input <your file>

# ex:
$ pfnstudio run timesfm --input sample

# download it without running · its tasks, inputs, license and digest
$ pfnstudio pull timesfm
$ pfnstudio show timesfm

# models loaded now · delete it from this machine
$ pfnstudio ps
$ pfnstudio rm timesfm

Python 3.10+. The first run downloads the model's adapter, asks before installing PyTorch and the model's own packages (-y installs without asking) and pulls the pretrained weights from Hugging Face. --input sample runs the sample input pfnstudio ships and prints its path: copy it, put your data in, and pass it as <your file>. Later runs reuse the cache.

Serves

Headline numbers

Provenance

Readme

A ready-to-use forecasting model you can point at your own history — it predicts the next steps of sales, demand or any metric, with a range, without training a model.

TimesFM 2.5 (Google Research, arXiv:2310.10688 lineage) is a decoder-style patch transformer (~200-230M parameters) for zero-shot probabilistic time-series forecasting. It reads a series' history (up to ~1024 steps of context) and returns a point forecast plus deciles for up to ~256 steps ahead, in one forward pass. This listing installs the real pretrained weights as an inference-only base model: the checkpoint is pulled from HuggingFace (google/timesfm-2.5-200m-pytorch) and served through the upstream timesfm package. Continued pretraining is not offered — the upstream release ships no training prior.