rt_seen43(job 112772, own qos 8 GPU, 07h 경과/36h 한도, 2.16 s/it → ETA ≈19h) 진행 중. step 76 기준 latent_loss ≈0.25 / action_loss ≈0.023. 논문 baseline(LingBot-VA unseen 7 평균 17.45%)이 직접 비교 앵커 — 완료 후 hold-out 7 task 평가 배선(경로 2줄 패치) 착수 필요.evaluation/robotwin/eval_polict_client_openpi.py 경로 2줄 패치(구 RoboTwin 레이아웃 → 허브 레이아웃) + hold-out 7 task 러너 구현.RoboTwin hub의 rt_hub_flexpi 평가(job 112482)는 flex-pi 모델을 대상으로 하지만 RoboTwin 레포 자체의 작업이라 RoboTwin 카드에 반영했다 — flex-pi 저장소 자체의 커밋·claude 문서는 이번 기간 갱신 없음.
기간 내 non-batch job 104건(COMPLETED 69 · PREEMPTED 18 · CANCELLED 15 · RUNNING 2 · PENDING 0). 아래는 프로젝트별 대표/핵심 작업만 발췌.
| Job ID | 프로젝트 | 실행 명령 | GPU | 상태 | 실행 시간 | 노드 |
|---|---|---|---|---|---|---|
| lingbot-va | ||||||
| 112772 | lingbot-va | train_robotwin_seen43.sh — rt_seen43 (RoboTwin 43-seen 학습, own) | 8 | RUNNING | 07:01:28 | 1 |
| FurnitureBench | ||||||
| 112492 | FurnitureBench | fb_lingbot_train_all_e128p (packing 학습 완주, 3000/3000) | 8 | COMPLETED | 08:20:23 | 1 |
| 112911 | FurnitureBench | fb_eval_lingbot_e128p_one_leg_step3000 (30ep 최종평가) | 1 | RUNNING | 03:25:58 | 1 |
| 112912-914 | FurnitureBench | fb_eval_lingbot_e128p_{lamp,round_table,cabinet}_step3000 (3잡) | 1×3 | COMPLETED | ~1.5h | 1 |
| 112579-582 | FurnitureBench | fb_eval_lingbot_e128p_{one_leg,lamp,round_table,cabinet}_step2000 (4잡) | 1×4 | COMPLETED | ~1.5h | 1 |
| 112478-481 | FurnitureBench | fb_eval_lingbot_all_{one_leg,lamp,round_table,cabinet}_step7500 (e40 최종평가, 4잡) | 1×4 | COMPLETED | ~1.3-1.6h | 1 |
| 112276 | FurnitureBench | fb_lingbot_train_all_e128p (packing 학습 시작) | 8 | PREEMPTED | 02:06:12 | 1 |
| 112272 | FurnitureBench | fb_lingbot_train_all (e40, 7500에서 사용자 종료) | 4 | CANCELLED | 02:46:25 | 1 |
| 112223 / 112167 | FurnitureBench | fb_lingbot_train_all (e40, preempt 반복) | 4×2 | PREEMPTED | ~0.4-0.8h | 1 |
| 112137 | FurnitureBench | fb_lingbot_train_all_e128 (구 e128, 사용자 종료) | 8 | CANCELLED | 00:00:00 | 1 |
| scene_bench (scene-mem-gen 포함) | ||||||
| 112009-011 | scene_bench | run_v1_final_shards.sh — v11s2_{flash37,sonnet,luna} (Restore S② delta, API) | 0(CPU) | COMPLETED | ~0.5-1.6h | 1 |
| 112006-008 | scene_bench | v11s2open_{qwen3vl8b,cosmosr2,molmo2} (Restore S② delta, 오픈소스 vLLM) | 1×3 | COMPLETED | ~10-34m | 1 |
| 111838 | scene_bench | archive_final_results.py --family restore --images | 0(CPU) | COMPLETED | 00:02:53 | 1 |
| 112383 | scene_bench | run_routine_pilot.sh — routine_pilot (train_1652__3000, 5전략) | 0(CPU) | COMPLETED | 00:08:27 | 1 |
| 112438 | scene_bench | run_routine_pilot.sh — routine_twins (쌍둥이 2000/2001) | 0(CPU) | COMPLETED | 00:20:54 | 1 |
| 112200-203 | scene_bench | online_pilot4_v1final_{find_s0,find_s1,restore10,restore11} (검증 런) | 1×4 | COMPLETED | ~1.4-1.6h | 1 |
| 112171-174 / 112186-189 | scene_bench | online_pilot3_v1final_* (njoints 버그로 첫 제출 실패 → 재제출) | 1×8 | 혼합(일부 CANCELLED) | ~0.1-0.15h | 1 |
| 112423-430 / 112483-490 | scene_bench | online_mid_find_s0-s7 (500플랜 중간 런, $31, 재제출 포함) | 1×16 | 혼합(다수 COMPLETED, 일부 CANCELLED/PREEMPTED) | ~0.3-3.2h | 1 |
| 112526 / 112544-545 | scene-mem-gen | rt_s1_5455 / rt_s3_{4706,3118} (routine batch datagen) | 1×3 | COMPLETED | ~9m-34m | 1 |
| MolmoBot | ||||||
| 111202 | MolmoBot | train_pi05_mlspaces.sh — pi05_place_oc (no-goal 기준선, own) | 8 | COMPLETED | 11:52:19 | 1 |
| 111262 | MolmoBot | train_molmobot_goal.sh — mbot_place_oc (no-goal 기준선, own) | 8 | COMPLETED | 07:58:46 | 1 |
| 111733-739 | MolmoBot | rollout_handoff.sh / run_place_val.sh / rollout_val.sh — ho_ng20k 등 (no-goal eval, 7잡) | 1×7 | COMPLETED | ~0.7-4.2h | 1 |
| 111936-939 | MolmoBot | rollout_handoff_molmobot.sh / rollout_val_molmobot.sh — ho_mb_ng 등 (4잡) | 1×4 | COMPLETED | ~4-4.7h | 1 |
| RoboTwin | ||||||
| 112267 | RoboTwin | convert_legacy_worker.sh — rt_hub_convert | 0(CPU) | COMPLETED | 00:03:24 | 1 |
| 112357 / 112368 | RoboTwin | smoke_sim_worker.sh — rt_hub_smoke (HoldPose 스모크, 2잡) | 1×2 | COMPLETED | ~0.5-6.0m | 1 |
| 112379 | RoboTwin | eval_multitask_worker.sh — rt_hub_xwam (첫 시도, 사용자 종료) | 2 | CANCELLED | 00:44:53 | 1 |
| 112436 | RoboTwin | eval_multitask_worker.sh — rt_hub_xwam (X_WAM_RoboTwin 4 task, 97.5%) | 2 | COMPLETED | 00:59:31 | 1 |
| 112482 | RoboTwin | eval_multitask_worker.sh — rt_hub_flexpi (FlexPi 2 task, 100%/97%) | 2 | COMPLETED | 00:15:32 | 1 |