Date: March 26th Company: SAS Contact: Anand Surana <Anand.Surana@sas.com> Hi Team, Anand this side! I was trying to do time series forecasting for one of my use cases. Used the timegpt.forecast method to forecast for future periods. Although the code seems to be working fine facing issues when I try to include below parameters: • model='timegpt-1-long-horizon' • _*finetune_loss='mae'*_ _*If I remove these parameters from my method call, the TimeGPT tends to give a forecast but to finetune was trying with these 2 additional parameters and the code fails. Have added the snippet of code and error for your reference. Also, I am referring to https://docs.nixtla.io/docs/forecasting_on_a_long_horizon for the syntax and features of timeGPT.*_ Can you please let me know if there’s something which I am missing, resulting in the failure of code execution.

This has to do with the change of the parameters' name that Azul shared with us I think. It should be

model="long-horizon"

and

fewshot_loss="mae"

, I'll check the PR today to see if we can review and merge it quickly so the info gets updated in the documentation

Hello Anand, Thanks for using TimeGPT!. We recently made some changes to the SDK. You need to set

model='long-horizon'

and

fewshot_loss='mae'

. We will update the documentation soon.