GET
cURL
The schema below documents the non-ready response branches (202 / 409 / 500) only. When status='ready' this endpoint streams application/x-ndjson instead, with one DeepSeek-R1-shape row per line. See Training Datasets concept page for the full NDJSON row shape and the canonical token reference.

Worked example

A single NDJSON line from a successful export looks like this:
The <no_response/> token is emitted whenever the operator deliberately chose silence at a reach-out moment. It is a first-class learnable target, not an absence of data. See the token reference for the full table.

Polling pattern

Response headers on the success branch:

Authorizations

Authorization
string
header
required

API key obtained from Labs Portal

Path Parameters

dataset_id
string<uuid>
required

Training dataset UUID

Pattern: ^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$

Response

Success

status
enum<string>
required

Lifecycle status. The 200 NDJSON path is only served when status=ready.

Available options:
pending,
preview,
running,
filtering,
cancelled,
failed
row_count
integer | null
required

Completed rows so far. Null until finalised.

Required range: -9007199254740991 <= x <= 9007199254740991
message
string
required

Human-readable explanation

error
string

Stored error string when status=failed