CLI
reflex training
Start, watch, list, and stop fine-tuning runs.
Training runs operate on a validated dataset and produce a model artifact you can later deploy or use for inference.
reflex training start
Kick off a training run.
reflex training start --dataset ds_abc123| Flag | Default | Description |
|---|---|---|
--dataset | required | Dataset ID from reflex datasets register. |
--base-model | pi0.5 | Base model to fine-tune. Only pi0.5 is currently supported. |
--trainable | lora | lora or full. |
--epochs | 3 | Number of training epochs. |
--model-name | auto | Name for the resulting artifact. |
--model-version | auto | Version label, e.g. v1. |
--parameters-json | — | JSON literal or @path/to/file.json for extra hyperparameters. |
--request-id | — | Idempotency key. Re-running with the same ID returns the existing run instead of creating a new one. |
--json | off | JSON output. |
Passing hyperparameters from a file:
reflex training start \
--dataset ds_abc123 \
--trainable lora \
--parameters-json @params.jsonSample response:
{
"ok": true,
"trainingRun": {
"id": "tr_...",
"status": "queued",
"datasetId": "ds_abc123",
"modelName": "...",
"modelVersion": "..."
},
"next": ["reflex training status <training-run-id> --watch"]
}reflex training status
Check the status of a run.
reflex training status tr_abc123
reflex training status tr_abc123 --watch| Flag | Default | Description |
|---|---|---|
--watch | off | Poll until the run reaches a terminal status. |
--poll-interval | 10.0 | Seconds between polls when --watch is set. |
--json | off | JSON output. |
Terminal statuses: succeeded, failed, stopped.
reflex training list
List training runs in your organization.
reflex training list
reflex training list --allBy default this returns active runs plus runs that failed within the last 15
minutes. Pass --all to include the full history.
reflex training stop
Halt a running training run.
reflex training stop tr_abc123The run transitions through stopping to stopped. Any artifact already
checkpointed remains available.