Your First Program

A minimal Spark program submits a train job (dry fixtures), polls status, then asserts with expect — the train→status→expect loop you can put in a PR (exit 0 pass / exit 1 fail).

train_eval.spark

Create a file named train_eval.spark (or use examples/train_eval.spark):

model train dataset "examples/fixtures/train/dataset.jsonl" base "fixture-base" out "out/train/job-dry-001" backend "http" -> job

model status "job-dry-001" -> status

expect contains job fixture "examples/fixtures/train/want_accepted.txt"
expect contains status fixture "examples/fixtures/train/want_succeeded.txt"

backend "http" means Spark POSTs to a trainer you run at SPARK_TRAIN_URL. Dry-run never hits the network.

Run it

./spark --dry-run examples/train_eval.spark
# exit 0 — [expect] pass …

./spark --dry-run examples/train_eval_fail.spark ; echo $?
# → 1

Dry-run returns fixture job + status so you can iterate without a trainer or API key. For a live train submit (your service):

export SPARK_TRAIN_URL=http://127.0.0.1:8090/v1
./spark --live examples/model_train.spark

Add classify or more eval helpers in AI in 5 Minutes, then the Build a Model wizard. Optional freeform ask still works — see examples/hello.spark — but the product thesis is train → status → expect (spark_distill_cpu on the reference trainer).

Next: AI in 5 Minutes →