GitHub radar
TimesFM 3.0: Google's Forecasting Model Runs on a Laptop
Google Research released TimesFM 3.0, a small pretrained model that continues time series such as sales or traffic — weights are for non-commercial use.
TimesFM 3.0 is the latest version of Google Research's pretrained time-series foundation model for forecasting. You give it a history of values — daily revenue, site traffic, stock levels — and it returns the continuation of the series together with quantile ranges, so you see not one number but a spread of likely outcomes. It was pretrained on large collections of time series, including Wikipedia pageviews and Google Trends queries, plus synthetic data. This repository holds the official PyTorch weights; the model has about 330 million parameters. The weights are released under the TimesFM Non-Commercial License v1.0.
Why a vibe-coder should care
Forecasting used to mean either a spreadsheet trend line or hiring someone who builds models. With a small pretrained model, an AI agent can take your own numbers and produce a forecast with a range in minutes — useful for planning stock, budgets or content, as long as the use is non-commercial.
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.
Обычный ноутбук, видеокарта не нужна — модель занимает около 0.7 ГБ.
Open on Hugging Face▌ More finds