autotrain

name

autotrain

description

Operate the OpenUI SLM training pipeline end to end, including a continuous hands-off model and harness improvement loop. Bare /autotrain is non-terminating and must not stop for user confirmation; an explicit phase or --once is finite. Code fixes during training use incremental commits every cycle and stacked PRs only after positive-result runs (sdlc autotrain-iteration-delivery); when training stops, full bottom-up SDLC closeout of open positive layers.

Autotrain OpenUI SLMs

Facade for operating and continuously improving the training pipeline with progressive disclosure: this file routes; each phase's full instructions live in references/ and are read only when that phase is being run. To modify a harness, use improve-openui-harnesses. For the higher-level, knowledge-driven research loop that coordinates this pipeline with brains / OpenWiki / literature discovery / Linear, use autoresearch.

Delivery process for code and durable docs during training is owned by sdlc — read ../sdlc/references/autotrain-iteration-delivery.md before continuous or multi-run work.

Workflow

1. Bare /autotrain (default) is continuous and hands-off. Immediately enter references/continuous.md. Keep chaining bounded campaigns until the session is preempted or the repeated-blocker rule fires. Do not stop to ask the user to continue. Do not end a turn with only a resume recipe. Print the result matrix between cycles, then start the next cycle without waiting.
2. An explicit phase name or --once performs one finite pass only.
3. For finite phase work: pick the phase from the routing table below (slm list / slm guide <slug>), read references/<slug>.md plus references/contracts.md once per session, run the slm commands, and close out docs/model-card duties.
4. Hand off (when those claims appear): ship → honest-ship-eval; matrices → running-experiment-matrices; campaigns → openui-autoresearch; Lean bands → improve-lean-optimums; brains/OpenWiki/Linear → autoresearch; multi-layer land → sdlc.

Continuous mode (bare /autotrain) — non-negotiable

Rule	Required behavior
Hands-off	No confirmation prompts; no “say continue”
Non-terminating	An unbudgeted host goal owns persistence; each bounded cycle returns control before cycle N+1
Self-heal	Fix path/knob/harness failures from evidence; re-run
Soft failures	Fixture ship-gate fails / null deltas / single timeouts → next cycle
Hard block only	Same unrecoverable blocker 3× with no new info → report blocked
Incremental commits	Commit green code/docs units every cycle while working on an iteration or fix
Stacked PR (positive only)	Open/update a gh stack layer only after a positive-result run (metric win, ship-quality win, or proven executable unblock)
Non-positive cycles	Docs + local commits only — no new stack layer for fixture fails / null deltas
Get latest between runs	git fetch + gh stack sync / merge origin/main; resolve conflicts
Remote compute default	No paid GPU / HF write without prior user authority
Training stopped	Full sdlc bottom-up closeout of open positive layers (review → CI → squash-merge) — not a resume paste
Matrix to the user	After every cycle (and whenever reporting status): paste the compact four-table view (liveness, results, diagnostics, priorities) from status --loop-id <id> --matrix --last 5 (or /tmp/autotrain-report.sh / /tmp/autotrain-loop-dashboard.md) into the chat. Never claim progress without it
Liveness proof	Prove state with the host goal plus loops/<loop-id>/state.json heartbeat, phase, next action, and child PID. Legacy drivers may also use /tmp/autotrain-loop-status.txt. Never use Grok “background ops” UI as liveness
Never kill the loop	Do not kill/pkill/kill -9 run_autotrain_continuous, its children, or the continuous worktree processes to ship skills, fix CI, merge PRs, or “restart cleanly.” Side work uses another worktree/branch

After each supervised cycle, validate cycle_handoff.json and execute its typed actions in order: theorem stop → harness repair → formal repair → data repair → docs/delivery → successor experiment. The handoff owner names the required skill; do not reduce it to a prose recommendation.

Full procedure: references/continuous.md.
Delivery: ../sdlc/references/autotrain-iteration-delivery.md.
User-facing report (preferred): bash /tmp/autotrain-report.sh → updates /tmp/autotrain-loop-dashboard.md.

Phase routing

Phase	Command	Reference
Continuous model + harness improvement (default)	keep chaining campaigns; status --loop-id <id> --matrix between cycles	references/continuous.md
Build/publish training corpora	slm data build-train / publish-train / store	references/train-data.md
Build held-out/adversarial/OOD suites	slm data build-test	references/test-data.md
SFT / model build (Phase A)	slm sft train / remote / hf-jobs	references/sft.md
Evaluate + ship gates	slm eval model / diagnose / loss-suites / tasks	references/eval.md
Distillation / P1–P3 climb	slm distill collect / self / resume-climb	references/distill.md
Preference / surrogate-DPO (Phase B)	slm preference <subcommand>	references/preference.md
RL / GRPO-lite (Phase C; NeMo/MOLT)	slm rl train / nemo / molt	references/rl.md
Experiment matrices, scaling, recipes	slm experiments <matrix>	references/experiments.md
Checkpoint sync, lineage, promotion	slm checkpoints sync / migrate; slm cycle <sub>	references/checkpoints.md
Annotation export → preference inputs	slm annotations export

Non-negotiable contracts

Digest — full versions in references/contracts.md:

• Iron law: no run without the matching docs/design/ JSON + markdown.
• Model card: every checkpoint updates docs/MODEL_CARD.md + README summary.
• Honesty: fixture/scratch evidence is wiring only; readiness needs --ship-gates on full scoreboards.
• RL is fail-closed: approved RLReadinessReport or no RL — no override.
• No shadow paths: reuse canonical scripts/harnesses and artifact roots.
• Decode invariants: constrained decoding is the product. Deterministic singleton bypass outranks any learned score; unconstrained arms are diagnostic controls, never defaults, serving paths, or gate inputs. Full law in AGENTS.md §Non-negotiable architecture invariants + decode-invariants.md.