refactor: round 2 — runstate markers, resolve_spec, versioned supervision - #71
Merged
Conversation
…sion (R1/R2/R4) R1 runstate: the marker protocol (best/config.json, final_eval.json, chain_log.jsonl, step-N) gets an interface — RunState, mkdir-safe by construction (the arm-4 bug class is unrepresentable). The orchestrator consumes it; pure-CPU tests. R2 supervision as code: scripts/modal_retry.sh + scripts/watch_marker.sh replace the seven bespoke /tmp loops this campaign produced. R4 resolve_spec: the four pasted vol_map blocks collapse into one owner of 'which volume does this spec's teacher/student/dataset live on', verified against real specs (rababa/secrets/persian mounts, hub teachers, data_volume overrides).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Round-2 deepening (report delivered alongside):
R1 — RunState (STRONG). The marker protocol (
best/config.json,final_eval.json,chain_log.jsonl,step-N) becomes an interface:RunStatewithtraining_done()/eval_done()/latest_step()/log()/read_eval(), mkdir-safe by construction — the arm-4 orchestrator bug (first log into a not-yet-created run dir) is unrepresentable through it. The orchestrator consumes it. Pure-CPU tests cover the exact incident semantics.R2 — supervision as versioned code (STRONG).
scripts/modal_retry.sh(the generic until-loop) +scripts/watch_marker.sh(volume-marker watcher) — this campaign wrote seven bespoke /tmp loops for the same two patterns; two died silently with the workstation.R4 — resolve_spec. Four pasted
vol_mapblocks (distill, evaluate, evaluate_der + final_eval write, microkimi/probe) collapse into one owner of volume-layout knowledge; verified against real specs (rababa/secryst/persian mounts, hub teachers,data_volumeoverrides). R3 (rababa harness vendoring + the provenance ADR) is rababa PR #58.Suite green (runstate + harness tests; ruff clean on touched files).