fix: orchestrator log() crashes on a fresh arm's missing run dir - #67
Merged
Conversation
added 6 commits
August 28, 2026 12:52
…ard to 2.2864 WikiNews-2024 multi-ref: 17.3794/11.8273 vs r6's 19.8191/12.4613. In-domain 2.2864 vs 2.5793. The teacher lineage no longer trades ID for OOD. Future distillations take r7 as teacher.
ara-diac-2.0 = the r7 canonical teacher (ID 2.2864/1.3343, OOD 17.3794/11.8273 — full sweep over 1.0). Metadata + model card + export entry; provenance anchors in rababa RESULTS.md (r7 verdict tables). Export runs the new gates: int8 with fp32 head, CER parity + margin budget. ara-diac-small-2 = run-006-r7-muon: r7 teacher + Muon (the two measured wins compound). E4 pre-registered: gate ≤ 6.26 (≥2pp over the shipped 8.259), prediction 4.3-5.0. Orchestrator arm added; idempotent.
Optimizer alone −2.96pp, memory alone −0.70pp (−0.47 under Muon), combined −3.43pp — roughly additive. Optimization is the dominant recoverable term of the distillation gap; capacity second; ~2.2pp residual is domain coverage.
The first watch line for an arm whose training hasn't started opens chain_log.jsonl inside a directory that only distill_sequence creates — FileNotFoundError killed the orchestrator right after run-005's eval (exactly when it reached arm 4), and every relaunch died the same way before logging anything. mkdir(parents=True, exist_ok=True).
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
The qwen_next_chain orchestrator crashed the moment it reached a fresh arm whose training hadn't started:
log()openedchain_log.jsonlinside the arm's run directory, which onlydistill_sequencecreates. FileNotFoundError — no traceback surfaced because the function is a fire-and-forget spawn, so it died silently after run-005's eval (exactly when it advanced to arm 4 / run-006) and every relaunch repeated the death before logging anything.One-line fix:
mkdir(parents=True, exist_ok=True)inlog().This was the "why did the orchestrator die right after run-005's eval" mystery — not an eviction.