---
title: "Foundational Time Series Model Multi Series Finetuning"
method: POST
path: "/v2/finetune"
---

# Foundational Time Series Model Multi Series Finetuning

`POST /v2/finetune`

Fine-tune the large time model to your data and save it for later use. It takes a JSON as an input containing information like the series frequency and historical data. (See below for a full description of the parameters.) The response contains the ID of the finetuned model, which you can provide in other endpoints to use that model to make the forecasts. Get your token at https://nixtla.io/free-trial?utm_source=nixtla.io&utm_campaign=/docs/api-reference.

## Request body

- FinetuneInput
  - `series` SeriesWithFutureExogenous, required
    - `X_future` array[], nullable — Future values of the exogenous features. Each feature must be a list of size number of series times the forecast horizon (h).
      - union[]
        - union
          - number
          - string
    - `X` array[], nullable — Historic values of the exogenous features. Each feature must be a list of the same size as the target (y).
      - union[]
        - union
          - number
          - string
    - `categorical_exog` integer[], nullable — Zero-based indices of the columns in X that are categorical features.
    - `y` number[], required — Historic values of the target.
    - `sizes` integer[], required — Sizes of the individual series.
  - `freq` string, required — The frequency of the data represented as a string. 'D' for daily, 'M' for monthly, 'H' for hourly, and 'W' for weekly frequencies are available.
  - `model` string — Model to use as a string. Common options are (but not restricted to) `timegpt-1` and `timegpt-1-long-horizon.` Full options vary by different users. Contact support@nixtla.io for more information. We recommend using `timegpt-1-long-horizon` for forecasting if you want to predict more than one seasonal period given the frequency of your data.
  - `finetune_steps` integer — The number of tuning steps used to train the large time model on the data. Set this value to 0 for zero-shot inference, i.e., to make predictions without any further model tuning.
  - `finetune_loss` 'default' | 'mae' | 'mse' | 'rmse' | 'mape' | 'smape' | 'poisson' — The loss used to train the large time model on the data. Select from ['default', 'mae', 'mse', 'rmse', 'mape', 'smape', 'poisson']. It will only be used if finetune_steps larger than 0. Default is a robust loss function that is less sensitive to outliers.
  - `finetune_depth` 1 | 2 | 3 | 4 | 5 — The depth of the finetuning. Uses a scale from 1 to 5, where 1 means little finetuning, and 5 means that the entire model is finetuned. Note that this parameter is only effective for timegpt-1 and timegpt-1-long-horizon models, meanwhile it has no effect on the other models. By default, the value is set to 1.
  - `output_model_id` string, nullable — ID to assign to the finetuned model
  - `finetuned_model_id` string, nullable — ID of previously finetuned model
  - `hist_exog` integer[], nullable — Zero-based indices of the exogenous features to treat as historical.
  - `multivariate` boolean — Compute multivariate predictions across a batch of multiple time series. Requires all time series with overlapping dates. Note that this is only effective for timegpt-2.1 model and it has no effect on the other models. By default, the value is set to False.
  - `model_parameters` object, nullable — Optional dictionary of parameters to customize the behavior of the large time model.

## Response `200`

Successful Response

- FinetuneOutput
  - `input_tokens` integer, required
  - `output_tokens` integer, required
  - `finetune_tokens` integer, required
  - `finetuned_model_id` string, required

## Other responses

- `422` — Validation Error

---

[API](https://skmtc.net/nixtla/apis/nixtla-forecast-api.md) · [All operations](https://skmtc.net/nixtla/apis/nixtla-forecast-api/llms.txt) · [OpenAPI document](https://skmtc-service-staging.skmtc.workers.dev/v1/apis/nixtla/nixtla-forecast-api/revisions/a3b589142eba/schema)