GET https://api.fastino.ai/v1/datasets
X-API-Key: YOUR_API_KEY

Optional query parameters: include_all_versions, include_failed, include_orphans, project_id, limit, and offset. The list endpoint has no task_type filter.

Get a Dataset (Check It Exists)

Returns all versions of the dataset:

GET https://api.fastino.ai/v1/datasets/YOUR_DATASET_NAME
X-API-Key: YOUR_API_KEY

Returns 404 if the dataset does not exist. Use this to confirm a dataset is ready before starting training.

Dataset Row Format

Each line of your JSONL must be a single JSON object. The required columns depend on the dataset type passed as dataset_type on POST /v1/datasets/upload/url.

NER (dataset_type: "ner")

text (string) — input text
entities (list of [span, label] pairs) — each span must be a substring of text. Use [] for negative examples (rows that contain no entities).

{"text": "Ada Lovelace worked in London.", "entities": [["Ada Lovelace", "PERSON"], ["London", "LOCATION"]]}
{"text": "No tagged entities here.", "entities": []}

Classification (dataset_type: "classification")

text (string)
label (string) for single-label, OR labels (list of strings) for multi-label. Pick one and use it consistently across the entire dataset.

{"text": "Great product", "label": "positive"}
{"text": "Mixed feelings about this", "label": "neutral"}

Decoder (dataset_type: "decoder")

Config jobs don't accept decoder base models, so a decoder dataset can't start POST /v1/training/jobs. To train a decoder, tokenize the examples and run them through an interactive job created with POST /v1/training/jobs/create-model.

Public decoder uploads use messages with {role, content} objects in every row.

{"messages": [{"role": "user", "content": "What is 2+2?"}, {"role": "assistant", "content": "4"}]}

Fastino also accepts the GLiNER2 canonical wire shape on upload: {"input": ..., "output": {"entities": {"<label>": ["<span>", ...]}}}. Classification rows can put one or more tasks in output.classifications. Use either the public fields or the canonical fields in a row; do not mix both shapes.

Upload a Dataset

Upload in two steps: get a presigned URL, then process the uploaded file.

Step 1 — get presigned upload URL:

POST https://api.fastino.ai/v1/datasets/upload/url
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "dataset_name": "my-dataset",
  "filename": "data.jsonl",
  "dataset_type": "ner"
}

Returns presigned_url, dataset_id, and version_number. PUT the exact file bytes to presigned_url, then retain the ID and version for processing and status polling.

Step 2 — process after upload:

POST https://api.fastino.ai/v1/datasets/upload/process
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "dataset_id": "DATASET_ID_FROM_STEP_1"
}

Returns HTTP 202 immediately. Poll GET /v1/datasets/{name} and select the entry whose version_number exactly matches the upload response. Continue until its status is ready or failed. If it is failed, stop and surface its processing_error; do not submit that version to training.

Delete an Entire Dataset (All Versions)

DELETE https://api.fastino.ai/v1/datasets/YOUR_DATASET_NAME
X-API-Key: YOUR_API_KEY

Use a Dataset for Training

Pass the dataset name directly to POST /v1/training/jobs:

{
  "datasets": [{"name": "YOUR_DATASET_NAME"}]
}

See the fastino-fine-tune skill for the full training workflow.

Error Codes

Code	Meaning
401	Invalid or missing API key
402	Insufficient credits
404	Dataset not found
422	Synchronous request-shape error — check body fields before retrying
500	Server error — safe to retry

Dataset-content validation runs asynchronously after processing returns 202. Content errors appear as a matching version with status: "failed" and a processing_error, not necessarily as a synchronous 422.

fastino-datasets

Fastino Datasets

Authentication

List Datasets