Create or confirm your dataset exists (GET /v1/datasets/:name)
Start training with that exact dataset name
Poll until status is deployed
Use the id field from the training job response as model_id for inference

Start a Training Job

base_model is required. Use a model ID from GET /v1/base-models or a checkpoint UUID. datasets is an array — supports multi-dataset training.

POST https://api.fastino.ai/v1/training-jobs
X-API-Key: YOUR_API_KEY
Content-Type: application/json

{
  "model_name": "my-ner-model",
  "base_model": "fastino/gliner2-base-v1",
  "datasets": [{"name": "YOUR_DATASET_NAME"}],
  "training_type": "lora",
  "nr_epochs": 5,
  "learning_rate": 5e-5
}

Key Response Fields

POST /v1/training-jobs returns:

id — the training job ID; this is your model_id for inference
status — current job status (see values below)

Poll Training Status

GET https://api.fastino.ai/v1/training-jobs/YOUR_TRAINING_JOB_ID
X-API-Key: YOUR_API_KEY

Status	Meaning
`requested`	Queued, waiting for GPU
`running`	GPU training in progress
`complete`	GPU training finished, post-processing starting
`normalizing`	Artifacts being normalized and deployed to inference provider
`deployed`	Ready for inference — use the job `id` as `model_id`
`errored`	Training failed — check logs
`stopped`	Manually stopped
`terminated`	Force-terminated

Poll until deployed before running inference.

Stream Training Logs

GET https://api.fastino.ai/v1/training-jobs/YOUR_TRAINING_JOB_ID/logs
X-API-Key: YOUR_API_KEY

Other Useful Endpoints

# List all jobs
GET https://api.fastino.ai/v1/training-jobs

# List checkpoints
GET https://api.fastino.ai/v1/training-jobs/YOUR_TRAINING_JOB_ID/checkpoints

# Download trained model
GET https://api.fastino.ai/v1/training-jobs/YOUR_TRAINING_JOB_ID/download

Stopping vs Terminating a Job

These are two distinct operations with different consequences:

Action	Endpoint	Effect	Reversible?
Stop	`POST /v1/training-jobs/{id}/stop`	Graceful stop; all checkpoints preserved; status → `stopped`	Yes — checkpoints remain usable
Terminate	`POST /v1/training-jobs/{id}/terminate`	Force-stops the job and deletes all checkpoints and artifacts; status → `terminated`	No — irreversible

# Stop gracefully (checkpoints preserved)
POST https://api.fastino.ai/v1/training-jobs/YOUR_TRAINING_JOB_ID/stop
X-API-Key: YOUR_API_KEY

# Terminate and delete all artifacts (irreversible)
POST https://api.fastino.ai/v1/training-jobs/YOUR_TRAINING_JOB_ID/terminate
X-API-Key: YOUR_API_KEY

Prefer stop unless you explicitly want to destroy all artifacts.

Either call may return 202 while the provider confirms the cancel: stop returns status: "stopping" and the job stays running until it flips to stopped; terminate flips to terminated and deletes checkpoints on confirmation. Poll GET /v1/training-jobs/{id} instead of calling again. A 409 with error.code training_job_already_finished means the job completed or failed first.

After Training

Once deployed, use the id from the response as model in /v1/chat/completions or POST /v1/gliner-2 for encoder tasks. See the fastino-inference skill.

Error Codes

Code	Meaning
401	Invalid or missing API key
402	Insufficient credits
404	Resource not found
409	`duplicate_training_submission` — an identical job was created moments ago and is still active; poll the `job_id` in the error instead of resubmitting. `idempotency_key_job_deleted` — the job this `Idempotency-Key` created was deleted; send a new key to launch it again
422	Validation error — check request body fields; `idempotency_key_reused` means the `Idempotency-Key` was already used for a different body
500	Server error — retry `POST /v1/training-jobs` with the same `Idempotency-Key`

Send an Idempotency-Key header (a fresh UUID per job you mean to create) on POST /v1/training-jobs and reuse it on every retry of that create: a repeat returns the original job instead of launching, and billing, a second run.

fastino-fine-tune

Fastino Fine-Tune

Authentication

Recommended Order