mrkeyoor.comAsk

google-research/timesfmtested 03 Oct 2026commit e51928e

timesfm review

TimesFM 3.0 is a candidate for a forecasting evaluation when its documented inputs fit your task and its weight license permits your use.

  • Install 31 s
  • Build 7 s
26/37passed, run exited non-zero
  • 54 packages
Dataevaluationupdated 26 Oct 2026
+45stars / 7d

TimesFM (Time Series Foundation Model) is Google Research's pretrained forecasting model and Python package. The README documents zero-shot univariate and multivariate forecasting in TimesFM 3.0, with past-only or past-and-future covariates and point forecasts plus nine quantiles. Downloaded 3.0 weights are restricted to non-commercial, non-production use.

Verdict

TimesFM 3.0 is a candidate for a forecasting evaluation when its documented inputs fit your task and its weight license permits your use. Our October 3, 2026 run at e51928e recorded successful install and build steps, but test collection encountered missing dependencies, and no forecasting workload was measured. Those errors do not establish that TimesFM is broken. Resolve the environment, inspect the nine recorded dependency advisories with coverage still unknown, and run dated backtests before making a deployment decision. Commercial or production users need an authorized route rather than assuming the repository's Apache-2.0 code license covers downloaded 3.0 weights.

We ran it

Lab card: what happened when we ran timesfm
Lab card from our sandbox run
Screenshot of timesfm (research.google/blog/a-decoder-only-foundation-model-for-time-series-forecasting)
research.google/blog/a-decoder-only-foundation-model-for-time-series-forecasting
Install✓ · 31s54 packages · 113 MB
Build✓ · 7s
Tests✗ · 10s26 passed · 0 failed · 36 skipped · 11 errors of 37 (pytest)
Known vulns9(pip-audit)
Repo156 files~22,794 lines of source · 4.5 MB · 2 CI workflows · tests dir

Answers from our run

Does timesfm build from source?

Dependencies installed in 31 seconds (54 packages), and the build succeeded in 7 seconds. We cloned commit e51928e into a clean Debian container with 3 CPUs and no project-specific setup.

Do timesfm's tests pass?

No. The project's own test command finished in 10 seconds and reported 26 passed, 0 failed, 36 skipped, 11 collection errors (pytest).

Does timesfm have known vulnerabilities in its dependencies?

pip-audit flagged 9 known advisories in the dependency tree at the time of our run.

Who should not use timesfm?

Teams seeking unrestricted commercial or production use of downloaded TimesFM 3.0 weights. The README directs those uses to authorized Google Cloud services, including BigQuery ML; the checkpoint license also describes obtaining a commercial license from Google.

What are the alternatives to timesfm?

Chronos, NeuralForecast. TimesFM 3.

Setup3/5The October 3, 2026 run at e51928e recorded successful 31-second install and 7-second build steps. Missing torch and pandas appear in test collection; backend readiness and inference were not verified.
Docs4/5The supplied README details installation extras, input and output shapes, covariates, MLX use and the separate weight license. This assesses the documentation supplied, not successful execution of its examples.
Community2/5The supplied issue listing shows user reports, but it does not establish maintainer response times, resolution quality or adoption. Popularity figures are not used as support-quality evidence.
Maturity2/5A v3.0.0 release, tests directory and two detected CI workflows provide development signals. Our run did not verify inference or production operation, and the README disclaims official Google product support.

Discussed on

  1. hnGoogle's 200M-parameter time-series foundation model with 16k context327 points
  2. hnTimesFM: Time Series Foundation Model for time-series forecasting317 points
  3. hnTimesFM (Time Series Foundation Model)7 points

Who it’s for

Researchers evaluating pretrained forecasting models for uses permitted by the checkpoint license.
Data scientists comparing zero-shot forecasts with seasonal baselines and their existing forecasting system.
Python users who can configure the documented PyTorch or Apple Silicon MLX backend and validate it on their own data.

Who it’s NOT for

Teams seeking unrestricted commercial or production use of downloaded TimesFM 3.0 weights. The README directs those uses to authorized Google Cloud services, including BigQuery ML; the checkpoint license also describes obtaining a commercial license from Google.
Teams requiring an officially supported Google product from this open release.
Buyers requiring demonstrated forecast accuracy, inference speed or deployment readiness from this review. Our recorded run measured none of those workloads.

Setup reality

In our recorded run on October 3, 2026, at commit e51928e27119cb17bebc005be2696b75e0a9e688, the harness marked its install step successful in 31 seconds and its build step successful in 7 seconds. It recorded 54 dependencies and 113 MB of dependency installation space. The install tail identifies Python 3.12.12; the recorded sandbox configuration was 3 CPUs, 8 GB of memory and image ghcr.io/astral-sh/uv:python3.12-bookworm.

The test step exited 1 after about 10 seconds. Its summary reported 26 passed, 36 skipped, zero failed and 11 errors. Visible exceptions identify missing torch in several current tests and missing pandas in two archived v1 tests. The supplied tail does not establish the cause of every error. The parser's total of 37 is not a verified exhaustive count of unique tests, so these results should not be expressed as a pass fraction.

Pip-audit recorded nine advisories across four packages. Audit coverage and the number of scanned packages are unknown; the record does not establish exploitability or identify fixes. Successful install and build step statuses do not establish that a forecasting backend or checkpoint was installed and usable. No installed-binary check, successful model inference, forecast-accuracy workload or inference-speed measurement is recorded.

The checkout measurements were 156 files, approximately 22,794 source lines and 4.5 MB, with two CI workflows, a tests directory and no Dockerfile. For local inference, the README documents Torch and MLX installation extras and the google/timesfm-3.0-pytorch checkpoint.

TimesFM 3.0: check the license before the backtest

TimesFM can provide a pretrained forecasting baseline, but the source-code license and the current weight license permit different uses. Google Research's README documents univariate and multivariate forecasting with past-only or past-and-future covariates, without per-task tuning. Its examples return point forecasts and nine quantiles, from 0.1 through 0.9. These are documented capabilities, not forecasting results from our run.

The repository source code is Apache-2.0, and the README says weights through version 2.5 retain that license. Downloaded TimesFM 3.0 weights are restricted to non-commercial, non-production use. The checkpoint license defines permitted research and evaluation narrowly: results must not be used in commercial decision-making, client deliverables or paid products and services. An internal prototype is not automatically permitted merely because it has not reached production.

The README permits commercial and production use through authorized Google Cloud services, including BigQuery ML. The checkpoint license also says other uses require a commercial license from Google, which it may grant at its discretion. The open release is not an officially supported Google product. Read the terms for your intended route before downloading weights or using their outputs.

What our recorded run establishes

Our stored run is dated October 3, 2026, at commit e51928e27119cb17bebc005be2696b75e0a9e688. The harness marked its install step successful in 31 seconds and its build step successful in 7 seconds. It recorded 54 dependencies and 113 MB of dependency installation space. The install tail identifies Python 3.12.12. The sandbox configuration records 3 CPUs, 8 GB of memory and image ghcr.io/astral-sh/uv:python3.12-bookworm.

These are command-step observations. The supplied record does not establish which optional backend was installed, that a checkpoint was downloaded, or that model inference worked. It contains no installed-binary check or measured forecasting workload.

The test step exited 1 after about 10 seconds. The visible pytest summary reports 26 passed, 36 skipped and 11 errors; the parser records zero failed. Several visible collection exceptions report missing torch, while two archived v1 test files report missing pandas. The tail does not expose the cause of every error, so it cannot support a single diagnosis for all 11.

Skipped tests are not failures, and collection errors do not establish incorrect forecasts. The parser's total field is 37, but exhaustive unique test coverage was not reconciled. Neither '26 of 32' nor '26 of 37' is a supported pass fraction. The useful conclusion is that this environment did not complete a clean test run and needs its dependencies checked before another attempt.

Pip-audit recorded nine advisories across four packages. The number of scanned packages and audit coverage are unknown. The record does not identify affected versions, fixes or reachable vulnerable behavior, so it supports follow-up dependency review rather than a claim that TimesFM itself is exploitable.

The checkout contained 156 files, approximately 22,794 source lines and 4.5 MB. The harness detected two CI workflows and a tests directory, with no Dockerfile. These describe the checkout, not production readiness.

Configure a backend, then evaluate forecasts

The README documents separate Torch and MLX installation extras. Its TimesFM 3.0 PyTorch example selects a CUDA device and the google/timesfm-3.0-pytorch checkpoint. It also documents an MLX-native backend for Apple Silicon. We did not verify either inference path in the recorded run.

The documented API accepts batches with different history lengths. The README says contexts longer than 15,360 points are truncated to their most recent observations before decoding. Check that behavior against any task that depends on older seasonal history, and validate covariate alignment, forecast shapes and the requested horizon before comparing errors.

For a permitted evaluation, use rolling historical cutoffs and compare the same horizons against a seasonal baseline and your existing system. Examine important series groups separately rather than relying only on an aggregate error. This review supplies no accuracy or speed result that would justify replacing an existing forecaster.

Two comparison paths

Chronos is a relevant pretrained-model comparison. Its README documents Chronos-2 zero-shot support for univariate, multivariate and covariate-informed tasks. That makes it a candidate for the same evaluation dataset, without establishing a performance winner.

NeuralForecast provides a different workflow: its README documents a collection of neural forecasting models with fit and predict interfaces. Consider it when fitting models to your data is part of the intended comparison. Neither alternative was executed in our recorded run.

How to read the scores

The scores are editorial judgments about the supplied evidence, not benchmark results or an overall product ranking. On this five-point rubric, 1 means the evidence offers little basis for proceeding, 2 means limited support with major unknowns, 3 means enough support for a controlled evaluation, 4 means detailed support for the assessed area, and 5 requires broad corroboration within that area. Missing evidence limits confidence; it does not prove a product defect.

The README provides substantial setup and API detail, but our run leaves backend readiness and inference unverified. The supplied issue listing shows reports about examples, tests and model behavior; titles alone do not establish confirmed defects or maintainer response quality. Community support quality and production maturity therefore remain uncertain.

Start by choosing a license-compatible route. Then configure the documented backend, confirm checkpoint loading, resolve collection dependencies and run a reproducible backtest. Those steps answer the deployment question that successful install and build statuses alone cannot.

Alternatives

ProjectWhat it isPick it when
ChronosA package providing access to the Chronos family of pretrained time-series forecasting models.Choose it for a separate zero-shot model comparison: its README documents Chronos-2 support for univariate, multivariate and covariate-informed forecasting. This review does not establish which model is more accurate.
NeuralForecastA library of neural forecasting models with fit and predict interfaces.Choose it when you want to fit and compare neural forecasting models on your own data rather than limit the evaluation to a pretrained zero-shot model.

What people are saying

  1. [github-trending] google-research/timesfm

Sources

  1. TimesFM repository and README
  2. TimesFM v3.0.0 release
  3. TimesFM 3.0 checkpoint license
  4. TimesFM open issues

More data reviews