GitHub radar
TimesFM 3.0 : The Google Predictive Model on a Laptop
Small pretrained model to predict future values of time series like sales or traffic data. Non-commercial weights available.
TimesFM 3.0 is a new release of Google Research’s pretrained time-series foundation model for forecasting. Given a time series (e.g., daily sales, website traffic, or inventory), TimesFM predicts future values and quantiles, giving a distribution of possible values instead of just a single point estimate. The model was trained on many time series datasets, including Wikipedia pageviews and Google Trends searches, as well as synthetic data. Official PyTorch weights are here. It has ~330M parameters. Model weights licensed under TimesFM Non-Commercial License v1.0.
Why a vibe-coder should care
Previously, forecasting was the domain of a spreadsheet trendline or a human data scientist. Now, with small pretrained models, you can get predictions with uncertainty intervals in minutes, which could help you plan your inventory, budgets, or content (non-commercially).
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up Google's TimesFM 3.0 forecasting model from https://huggingface.co/google/timesfm-3.0-pytorch — take my daily data table, forecast the next month and show a chart with the value range. Keep in mind the license is non-commercial.
A regular laptop, no GPU needed — the model takes up about 0.7 GB.
Open on Hugging Face▌ More finds