Skip to Content

Datasets

Manage training and evaluation datasets with format validation.

Supported Formats

FormatExtensionUse Case
OpenAI JSONL.jsonlChat fine-tuning (messages array)
HuggingFace.jsonInstruction tuning (instruction/output pairs)
Raw JSON.jsonCustom formats

Creating Datasets

Via Console

  1. Go to /dashboard/datasets
  2. Click “Create Dataset”
  3. Enter name, description, and format
  4. Upload file or paste data

Via API

# Upload a dataset
dataset = model_svc.create_dataset({
    "name": "customer-support-v2",
    "description": "Support conversation examples",
    "format": "openai_jsonl"
})
 
# Upload data file
model_svc.upload_dataset_file(dataset["id"], open("data.jsonl", "rb"))

Format Validation

Before export, datasets are validated against the target format schema:

OpenAI JSONL requires:

{"messages": [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}

HuggingFace requires:

{"instruction": "...", "output": "..."}

Exporting

Export datasets in any supported format via Console or API:

data = model_svc.export_dataset(dataset_id, format="openai_jsonl")

The Console shows a preview of the first 3 rows in the selected format before exporting.

Home