queue status update

queue_status.json

@@ -1,9 +1,9 @@
 {
-  "timestamp": "2026-04-12 04:45:59",
+  "timestamp": "2026-04-12 04:47:59",
   "total": 98,
-  "pending": 82,
+  "pending": 81,
   "running": 6,
-  "done": 9,
+  "done": 10,
   "failed": 1,
   "stale": 0,
   "retrying": 0,
@@ -15,7 +15,7 @@
       "gpu": 0,
       "status": "done",
       "elapsed": 1786,
-      "idle_time": 8297,
+      "idle_time": 8417,
       "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": 7982,
+      "idle_time": 8102,
       "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": 5330,
+      "idle_time": 5450,
       "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": 7716,
+      "idle_time": 7836,
       "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": 3356,
+      "idle_time": 3476,
       "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": 6970,
+      "idle_time": 7090,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_005_add_sub_baseline_100K_gpu2.log"
@@ -86,8 +86,8 @@
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --nu",
       "gpu": 0,
       "status": "running",
-      "elapsed": 8290,
-      "idle_time": 38,
+      "elapsed": 8410,
+      "idle_time": 158,
       "exit_code": -1,
       "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": 3202,
+      "idle_time": 3322,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_037_as_sorl_abs10_K4_25K_gpu0.log"
@@ -530,8 +530,8 @@
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 50 --K 4 --num",
       "gpu": 0,
       "status": "running",
-      "elapsed": 3195,
-      "idle_time": 219,
+      "elapsed": 3315,
+      "idle_time": 0,
       "exit_code": -1,
       "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": 1366,
+      "idle_time": 1486,
       "exit_code": -9,
       "retries": 1,
       "log_file": "/tmp/gpu_queue/job_053_as_sorl_abs10_K4_zipf10.0_500K_gpu1.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": 2525,
-      "idle_time": 9,
+      "elapsed": 2645,
+      "idle_time": 13,
       "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": 2530,
+      "idle_time": 2650,
       "exit_code": 0,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
@@ -949,10 +949,10 @@
       "name": "as_baseline_500K_1L2H256d",
       "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 500000 --num_epochs 20 --n_l",
       "gpu": 2,
-      "status": "running",
-      "elapsed": 6965,
-      "idle_time": 80,
-      "exit_code": -1,
+      "status": "done",
+      "elapsed": 7064,
+      "idle_time": 23,
+      "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": 1382,
+      "idle_time": 1502,
       "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": 1365,
-      "idle_time": 1,
+      "elapsed": 1485,
+      "idle_time": 20,
       "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": 7975,
-      "idle_time": 6,
+      "elapsed": 8095,
+      "idle_time": 2,
       "exit_code": -1,
       "retries": 0,
       "log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
@@ -1152,13 +1152,13 @@
       "job_id": 95,
       "name": "as_sorl_abs10_K4_250K",
       "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 4 --nu",
-      "gpu": -1,
-      "status": "pending",
-      "elapsed": 0,
-      "idle_time": 0,
+      "gpu": 2,
+      "status": "running",
+      "elapsed": 20,
+      "idle_time": 12,
       "exit_code": -1,
       "retries": 0,
-      "log_file": ""
+      "log_file": "/tmp/gpu_queue/job_095_as_sorl_abs10_K4_250K_gpu2.log"
     },
     {
       "job_id": 96,