google-research/timesfm ? reverse-engineered prompt

Hi, I'd like to build a Python tool for time-series forecasting using Google's TimesFM model. The idea is to have a script where I can input some historical data, like a numpy array of daily sales figures, and get a forecast for a specified number of future days.

Please base this on the google-research/timesfm repository. It seems they have a few different model versions; let's use the latest recommended one, which I think is 2.5.

A key feature I want is not just a single predicted value for each future time step, but also a confidence range or quantile forecast—for example, the 10th through 90th percentile outcomes. This will help me understand the potential upside and downside risk.

You can set it up with a simple example that shows how to load the model, prepare some dummy data, and then generate both the point and quantile forecasts. Please handle all the necessary dependencies like PyTorch and any other packages mentioned in their setup instructions.