queue status update

queue_status.json

@@ -1,10 +1,10 @@
 {
-  "timestamp": "2026-04-12 06:22:31",
+  "timestamp": "2026-04-12 06:24:53",
   "total": 98,
-  "pending": 79,
+  "pending": 77,
   "running": 5,
-  "done": 13,
-  "failed": 1,
+  "done": 14,
+  "failed": 2,
   "stale": 0,
   "retrying": 0,
   "jobs": [
@@ -15,7 +15,7 @@
       "gpu": 0,
       "status": "done",
       "elapsed": 1786,
-      "idle_time": 14088,
+      "idle_time": 14231,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_000_add_sub_baseline_10K_gpu0.log"
@@ -27,7 +27,7 @@
       "gpu": 1,
       "status": "done",
       "elapsed": 2101,
-      "idle_time": 13773,
+      "idle_time": 13916,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_001_add_sub_baseline_25K_gpu1.log"
@@ -39,7 +39,7 @@
       "gpu": 2,
       "status": "done",
       "elapsed": 4753,
-      "idle_time": 11122,
+      "idle_time": 11264,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_002_as_sorl_abs10_K1_25K_gpu2.log"
@@ -51,7 +51,7 @@
       "gpu": 0,
       "status": "done",
       "elapsed": 2366,
-      "idle_time": 13508,
+      "idle_time": 13650,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_003_add_sub_baseline_50K_gpu0.log"
@@ -63,7 +63,7 @@
       "gpu": 1,
       "status": "done",
       "elapsed": 6727,
-      "idle_time": 9147,
+      "idle_time": 9290,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_004_as_sorl_abs10_K1_50K_gpu1.log"
@@ -75,7 +75,7 @@
       "gpu": 2,
       "status": "done",
       "elapsed": 3112,
-      "idle_time": 12762,
+      "idle_time": 12904,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_005_add_sub_baseline_100K_gpu2.log"
@@ -87,7 +87,7 @@
       "gpu": 0,
       "status": "done",
       "elapsed": 8856,
-      "idle_time": 5228,
+      "idle_time": 5371,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_006_as_sorl_abs10_K1_100K_gpu0.log"
@@ -459,7 +459,7 @@
       "gpu": 0,
       "status": "done",
       "elapsed": 4510,
-      "idle_time": 8993,
+      "idle_time": 9136,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_037_as_sorl_abs10_K4_25K_gpu0.log"
@@ -531,7 +531,7 @@
       "gpu": 0,
       "status": "done",
       "elapsed": 5510,
-      "idle_time": 3479,
+      "idle_time": 3622,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_043_as_sorl_abs50_K4_50K_gpu0.log"
@@ -651,7 +651,7 @@
       "gpu": 1,
       "status": "failed",
       "elapsed": 5,
-      "idle_time": 7158,
+      "idle_time": 7300,
       "exit_code": -9,
       "retries": 1,
       "log_file": "/tmp/gpu_queue/job_053_as_sorl_abs10_K4_zipf10.0_500K_gpu1.log"
@@ -672,13 +672,13 @@
       "job_id": 55,
       "name": "as_sorl_abs100_K1_zipf5.0_500K",
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 1 --a",
-      "gpu": -1,
-      "status": "pending",
-      "elapsed": 0,
+      "gpu": 0,
+      "status": "failed",
+      "elapsed": 5,
       "idle_time": 0,
-      "exit_code": -1,
-      "retries": 0,
-      "log_file": ""
+      "exit_code": -9,
+      "retries": 1,
+      "log_file": "/tmp/gpu_queue/job_055_as_sorl_abs100_K1_zipf5.0_500K_gpu0.log"
     },
     {
       "job_id": 56,
@@ -769,10 +769,10 @@
       "name": "as_sorl_abs10_K1_50K_1L3H510d",
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 10 --K 1 --num",
       "gpu": 0,
-      "status": "running",
-      "elapsed": 5222,
-      "idle_time": 84,
-      "exit_code": -1,
+      "status": "done",
+      "elapsed": 5352,
+      "idle_time": 14,
+      "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_063_as_sorl_abs10_K1_50K_1L3H510d_gpu0.log"
     },
@@ -842,8 +842,8 @@
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
       "gpu": 2,
       "status": "running",
-      "elapsed": 8317,
-      "idle_time": 21,
+      "elapsed": 8459,
+      "idle_time": 12,
       "exit_code": -1,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_069_as_sorl_abs10_K1_500K_1L3H510d_gpu2.log"
@@ -903,7 +903,7 @@
       "gpu": 2,
       "status": "done",
       "elapsed": 2797,
-      "idle_time": 8322,
+      "idle_time": 8464,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
@@ -924,13 +924,13 @@
       "job_id": 76,
       "name": "as_baseline_250K_1L2H256d",
       "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 250000 --num_epochs 20 --n_l",
-      "gpu": -1,
-      "status": "pending",
-      "elapsed": 0,
-      "idle_time": 0,
+      "gpu": 0,
+      "status": "running",
+      "elapsed": 139,
+      "idle_time": 27,
       "exit_code": -1,
       "retries": 0,
-      "log_file": ""
+      "log_file": "/tmp/gpu_queue/job_076_as_baseline_250K_1L2H256d_gpu0.log"
     },
     {
       "job_id": 77,
@@ -951,7 +951,7 @@
       "gpu": 2,
       "status": "done",
       "elapsed": 7064,
-      "idle_time": 5695,
+      "idle_time": 5837,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_078_as_baseline_500K_1L2H256d_gpu2.log"
@@ -975,7 +975,7 @@
       "gpu": 1,
       "status": "done",
       "elapsed": 1971,
-      "idle_time": 7173,
+      "idle_time": 7316,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_080_as_baseline_25K_2L1H128d_gpu1.log"
@@ -1058,8 +1058,8 @@
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 1 --nu",
       "gpu": 1,
       "status": "running",
-      "elapsed": 7157,
-      "idle_time": 3,
+      "elapsed": 7299,
+      "idle_time": 7,
       "exit_code": -1,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_087_as_sorl_abs10_K1_250K_2L1H128d_gpu1.log"
@@ -1082,8 +1082,8 @@
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
       "gpu": 1,
       "status": "running",
-      "elapsed": 13767,
-      "idle_time": 263,
+      "elapsed": 13909,
+      "idle_time": 13,
       "exit_code": -1,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
@@ -1119,7 +1119,7 @@
       "gpu": 0,
       "status": "done",
       "elapsed": 3477,
-      "idle_time": 2,
+      "idle_time": 145,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_092_as_sorl_abs30_K4_10K_gpu0.log"
@@ -1154,8 +1154,8 @@
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 4 --nu",
       "gpu": 2,
       "status": "running",
-      "elapsed": 5692,
-      "idle_time": 4,
+      "elapsed": 5834,
+      "idle_time": 8,
       "exit_code": -1,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_095_as_sorl_abs10_K4_250K_gpu2.log"