# google-research/timesfm

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.

Repository: https://github.com/google-research/timesfm
Canonical: https://ross.abutalabs.com/products/timesfm
Homepage: https://research.google/blog/a-decoder-only-foundation-model-for-time-series-forecasting/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-07-14T03:27:18+00:00

## Health v2 (maintenance only)
Score: 81/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 92, release rhythm 79, longevity 61
- inputs: {"age_days": 856, "days_push": 50, "days_rel": 62, "gap_med": 73, "n_releases_24m": 4}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 28273, forks 2761 (observed 2026-08-28T04:11:48.227840+00:00)

## What it is
TimesFM is a pretrained decoder-only foundation model from Google Research for time-series forecasting, offering zero-shot forecasts with up to 16k context length and optional quantile heads. It ships as a Python package with PyTorch and Flax implementations, covariate support, and LoRA fine-tuning examples.

## Use cases
- forecast demand from historical sales time series
- zero-shot time-series forecasting without training a model
- predict future values of a metric given past observations
- fine-tune a pretrained forecasting model with LoRA
- generate quantile forecasts for uncertainty estimation
- forecast financial or retail time series in Python

## When to choose
- you need strong out-of-the-box forecasts without per-dataset training
- you want long-context forecasting up to 16k time points
- you need probabilistic quantile forecasts
- you want a lightweight 200M-parameter model for inference

## When to avoid
- you need fully interpretable classical methods like ARIMA or ETS
- you require real-time streaming forecasts at very low latency
- you need multivariate modeling with complex cross-series dependencies
- you cannot use GPU acceleration for reasonable inference speed

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference, llm-training, data-science
- domain: machine-learning, time-series, data-science, large-language-models
- platform: python, cross-platform
- tags: time-series-forecasting, foundation-model, zero-shot-forecasting, decoder-only, quantile-forecasting, pretrained-model, huggingface, lora-finetuning, gpu

## Member repositories
- google-research/timesfm (main) score 81

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:11:48.227840+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T16:54:24.880831+00:00, confidence not recorded.
  - readme: https://github.com/google-research/timesfm (fetched 2026-08-28T04:11:48.227840+00:00, sha 26225486ea03)
  - homepage: https://research.google/blog/a-decoder-only-foundation-model-for-time-series-forecasting/ (fetched 2026-08-29T07:50:56.289988+00:00, sha 96ea19b8b438)
- Data as of 2026-08-30T08:39:29.467469+00:00.