timesfm-forecasting

1# TimesFM Forecasting
2 
3## Overview
4 
5TimesFM (Time Series Foundation Model) is a pretrained decoder-only foundation model
6developed by Google Research for time-series forecasting. It works **zero-shot** — feed it
7any univariate time series and it returns point forecasts with calibrated quantile
8prediction intervals, no training required.
9 
10This skill wraps TimesFM for safe, agent-friendly local inference. It includes a
11**mandatory preflight system checker** that verifies RAM, GPU memory, and disk space
12before the model is ever loaded so the agent never crashes a user's machine.
13 
14> **Key numbers**: TimesFM 2.5 uses 200M parameters (~800 MB on disk, ~1.5 GB in RAM on
15> CPU, ~1 GB VRAM on GPU). The archived v1/v2 500M-parameter model needs ~32 GB RAM.
16> Always run the system checker first.
17 
18## When to Use This Skill
19 
20Use this skill when:
21 
22- Forecasting **any univariate time series** (sales, demand, sensor, vitals, price, weather)
23- You need **zero-shot forecasting** without training a custom model
24- You want **probabilistic forecasts** with calibrated prediction intervals (quantiles)
25- You have time series of **any length** (the model handles 1–16,384 context points)
26- You need to **batch-forecast** hundreds or thousands of series efficiently
27- You want a **foundation model** approach instead of hand-tuning ARIMA/ETS parameters
28 
29Do **not** use this skill when:
30 
31- You need classical statistical models with coefficient interpretation → use `statsmodels`
32…