Commit Graph

431 Commits

Author SHA1 Message Date
iomgaa 0414948682 merge: question-gen v3 + preflight fixes + continuous concurrent gate 2026-07-17 06:02:39 -04:00
iomgaa 9bb09a44d4 feat: add Slurm gpu-sbatch training entry for 4090-server 2026-07-17 06:02:39 -04:00
iomgaa b0be1f1ae5 docs: gate speedup final review and delivery record 2026-07-17 05:38:44 -04:00
iomgaa 172b7a8ec4 merge: continuous concurrent gate speedup (algo #6 semantic revision, design v3) 2026-07-17 05:29:42 -04:00
iomgaa eb12006d38 fix: idempotent ladder_rank migration for legacy gate_evidence tables 2026-07-17 05:29:20 -04:00
iomgaa 1930ad32a4 chore: raise Redis cache TTL to 7 days 2026-07-17 04:52:34 -04:00
iomgaa 8958eee11b refactor: remove block-sequential gate path and gate_block knob (algo #6)
config/train_videomme.yaml 同时收录待入库的实验配置变更(run_id v2 /
concurrency 32 / batch_size 40)。tests/integration/test_v3_contract_e2e.py
的 run_id 断言按 Task 5 显式契约同步修正(原断言依赖旧隐式实例注入)。
2026-07-17 04:40:30 -04:00
iomgaa 0b839937df fix: predictions row carries arm run_id under shared gate_log; drain evolve gather on failure (algo #6) 2026-07-17 03:52:19 -04:00
iomgaa 23a64042fe feat: parallel evolve + continuous gate wiring in runner (algo #6) 2026-07-17 03:32:14 -04:00
iomgaa b3aba7c31d fix: escape all LIKE specials in step-row cleanup 2026-07-17 01:17:31 -04:00
iomgaa 16993ed362 style: consolidate inference stub, complete docstrings (algo #6) 2026-07-17 01:13:19 -04:00
iomgaa ea6bec5421 fix: clear gate-derived rows on step rerun (idempotency) 2026-07-17 00:57:53 -04:00
iomgaa 9e8a254fbb fix: gate orchestrator materialize leak + cancel-drain on abort (algo #6)
Codex 质量审 2 Critical:
- C001: 候选物化移入 try、成功一个登记一个,第 N 个题型物化失败时
  finally 仍清理前 N-1 个已建目录,不泄漏
- C002: gather 首异常(护栏 raise)后显式取消其余任务并排水,确保
  finally 删除候选目录时无在飞任务访问、事件循环无 pending task 警告;
  护栏中止整轮语义不变
- 新增 2 测试:部分物化失败清理 / 护栏 raise 取消收束不悬挂(wait_for 5s)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-17 00:45:47 -04:00
iomgaa e8b66f85ab feat: continuous concurrent gate orchestrator (algo #6)
- validate_skills_concurrent: 多题型全部臂共享题槽并发编排,发射序 =
  题型 round-robin × 阶梯序(base 先 cand 后),终态统一组装 outcome,
  verdict None(全 INFRA)保留 RuntimeError 语义
- gate_evidence 列 block_idx → ladder_rank(阶梯序号,0-based);旧块路径
  _build_evidence_rows 仅键名同步(值仍为块号)保持落库兼容
- 新增 3 项编排测试:乱序到达前缀有序性/双题型隔离/全 INFRA raise

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-17 00:36:01 -04:00
iomgaa c61a6dac84 docs: soften slot fairness claim, complete docstrings (algo #6) 2026-07-17 00:28:25 -04:00
iomgaa 30c1cf10c0 fix: gate slot cancel-safety + post-inference freeze discard (algo #6)
Codex 质量审 4 项:推理后二次冻结检查(τ 后 in-flight 结果整体丢弃)、
acquire 取消回滚(半持有 permit 自动归还)、BoundedSemaphore 防静默扩容、
补取消恢复与冻结丢弃两个回归测试。
2026-07-17 00:25:18 -04:00
iomgaa 232afd525b feat: gate unit-arm tasks with question-slot gate (algo #6) 2026-07-17 00:13:51 -04:00
iomgaa 0a8e1ad18b docs: complete four-part docstrings for gate prefix section (algo #6) 2026-07-17 00:06:35 -04:00
iomgaa 2ab8088c81 test: strengthen frozen-noop full-state assertion (algo #6) 2026-07-17 00:03:36 -04:00
iomgaa 1e92928d4e fix: lock all-INFRA contract, idempotency tests, tuple units (algo #6) 2026-07-17 00:00:56 -04:00
iomgaa 21c360bc87 feat: gate prefix-ordered consumption core (algo #6)
CE-Gate 语义修订获批:块序贯 → 阶梯序前缀逐对序贯。新增 GateSpec/_UnitSlot/
_GateRun 数据结构与 _advance_prefix 纯逻辑(乱序到达下统计严格按预声明阶梯序
消费,INFRA 剔除后重判防 continue 悬置,过线即冻结)。旧块路径共存,Task 6 删。
2026-07-16 23:44:32 -04:00
iomgaa b02db74237 docs: apply Codex plan review fixes (tail-INFRA rejudge, slot deadlock guard) 2026-07-16 23:30:56 -04:00
iomgaa 85ccad3b72 docs: implementation plan for continuous concurrent gate 2026-07-16 23:12:53 -04:00
iomgaa 0e8069af27 docs: gate speedup v3 - ladder-prefix consumption fixes arrival-order bias (Codex round 2) 2026-07-16 22:46:01 -04:00
iomgaa 53a8c4be02 docs: redesign gate speedup v2 (continuous concurrent gate, drop preload) 2026-07-16 22:24:28 -04:00
iomgaa 269e0d294c docs: design for gate speedup (preload BaselineCache + parallel arms) 2026-07-16 14:23:47 -04:00
iomgaa cc01d5ed62 fix: address Codex review of telemetry fix (mkdir degrade, close, degrade tests) 2026-07-16 08:53:50 -04:00
iomgaa 065c8ae1b9 fix: telemetry single persistent connection + lock (kill concurrent write lock) 2026-07-16 08:47:23 -04:00
iomgaa 766592d855 docs: design for telemetry SQLite concurrency fix (align HarnessLog pattern) 2026-07-16 08:47:23 -04:00
iomgaa 933de6251e chore: add FORCE flag to split script; align train config pool sizes to 0.4 split 2026-07-16 07:46:31 -04:00
iomgaa 5fdcd9b5aa feat: add Video-MME 900 training entry (config + self-contained script) 2026-07-16 07:34:14 -04:00
iomgaa a1c4e051e7 Merge branch 'feat/preflight-train-fixes': preflight training fixes (4 WPs, 21+ defects) 2026-07-16 07:25:50 -04:00
iomgaa 6a9b1ea188 docs: record preflight final review + integration fixes (I-3/I-4/M-2) 2026-07-16 07:21:10 -04:00
iomgaa 6911c83d66 refactor: share INFRA stop-reason set from core.evolution 2026-07-16 07:19:31 -04:00
iomgaa b3ba11c7a5 fix: count gate INFRA guard numerator by unit not record 2026-07-16 07:16:33 -04:00
iomgaa ee69721ea3 fix: filter frozen pools by requested task_types subset 2026-07-16 07:13:31 -04:00
iomgaa e1f08dcd3b fix: count units not questions in trainability pre-flight; fail-fast when all types filtered 2026-07-16 06:45:16 -04:00
iomgaa ef244c52bd chore: add trainable_min_units to train_ar30 config 2026-07-16 06:35:10 -04:00
iomgaa ed59c6091a test: accept cache_salt kwarg in AgentLoop mock LLM stub 2026-07-16 06:34:16 -04:00
iomgaa 9be3bdb8eb feat: inject run_id as cache_salt for per-epoch resampling (algo #10) 2026-07-16 06:33:09 -04:00
iomgaa 58a0203522 perf: dedup holdout four-way eval (baseline derive, best_hard memo) 2026-07-16 06:31:00 -04:00
iomgaa 77fd35830c fix: idempotent _run_step (DELETE stale) + checkpoint after gate save 2026-07-16 06:27:38 -04:00
iomgaa efbdeb1647 fix: route None/degraded diagnoses to lapse; abort on high degrade rate 2026-07-16 06:24:11 -04:00
iomgaa 65126feada fix: patch checks full target span + marker injection/integrity (algo #8) 2026-07-16 06:20:29 -04:00
iomgaa caea9150b3 feat: pre-flight filter of untrainable task types before gate 2026-07-16 06:16:02 -04:00
iomgaa c44f6010eb fix: early_stop patience counts epochs not steps 2026-07-16 06:09:11 -04:00
iomgaa 25918a73ff fix: normalize non-scalar prediction; harden predictions insert 2026-07-16 06:06:58 -04:00
iomgaa 7d02cded99 fix: wrap diagnosis run_log with StepsJsonRunLog (restore algo #7 traces) 2026-07-16 06:05:55 -04:00
iomgaa e1cac644c8 docs: address Codex review of WP3 plan (pools passthrough, holdout memo, DELETE table) 2026-07-16 06:03:17 -04:00
iomgaa 03337af8f8 fix: gate INFRA isolation edge cases (all-INFRA fail-loud, parse_error in guard) 2026-07-16 05:54:02 -04:00