Training

With your data and model parameters configured, you're ready to train the foundation model. This section covers how to launch training and how to diagnose issues if something goes wrong.

In this section

Guide	Description
Pre-flight checklist	Verify your configuration before launching training
Run Training	Launch the foundation model training pipeline
Troubleshooting	Diagnose failures and fix common issues

Pre-flight checklist

Data sources — all tables connected, joins defined, column overrides applied (Managing Data)
Date ranges & split — training, validation, and (optionally) test start dates set correctly in data_params.split (Select & Organize)
Training parameters — learning rate, epochs, batch size, and device list reviewed (Tuning & Scaling)

Run a smoke test first

Limit batches and optionally train on a subset of entities:

yaml

training_params:
  limit_train_batches: 5
  limit_val_batches: 5

# optional: sample ~10 % of entities via where_condition
data_sources:
  - type: event
    name: transactions
    where_condition: "customer_id % 10 = 0"
    ...

Remove the limits once the pipeline completes without errors.

All good? Head to Run Training to launch the pipeline.