InstantMLDocs

Datasets

Group runs by the dataset and environment fields inferred from their configs.

Open .md

Use the Datasets tab to spot whether a comparison mixes runs trained on different data sources.

Understand what the tab reads

Dataset rows are derived from the run summaries currently loaded in the dashboard. They are not a separate dataset registry.

The table checks these config and metadata keys in order and uses the first non-empty string as the group name:

  • config.dataset
  • config.dataset_name
  • config.data
  • config.env
  • metadata.dataset
  • metadata.environment

Read the dataset rows

Each row shows:

  • The dataset or environment name inferred from config or metadata.
  • The number of runs that reference it.
  • The seed values seen in those runs.
  • The selected dashboard metric's best value for that group.

Name datasets consistently

Put stable data identity into run config so runs group cleanly:

python
import instantml as im

run = im.init(
    project="finetune",
    config={
        "dataset": "support-tickets-v3",
        "split": "train-2026-05",
        "seed": 42,
    },
)

The tab is most useful when projects use consistent config names across runs.

Next steps