queue status update
Browse files- queue_status.json +26 -26
queue_status.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"timestamp": "2026-04-12 06:
|
| 3 |
"total": 98,
|
| 4 |
"pending": 79,
|
| 5 |
"running": 6,
|
|
@@ -15,7 +15,7 @@
|
|
| 15 |
"gpu": 0,
|
| 16 |
"status": "done",
|
| 17 |
"elapsed": 1786,
|
| 18 |
-
"idle_time":
|
| 19 |
"exit_code": 0,
|
| 20 |
"retries": 0,
|
| 21 |
"log_file": "/tmp/gpu_queue/job_000_add_sub_baseline_10K_gpu0.log"
|
|
@@ -27,7 +27,7 @@
|
|
| 27 |
"gpu": 1,
|
| 28 |
"status": "done",
|
| 29 |
"elapsed": 2101,
|
| 30 |
-
"idle_time":
|
| 31 |
"exit_code": 0,
|
| 32 |
"retries": 0,
|
| 33 |
"log_file": "/tmp/gpu_queue/job_001_add_sub_baseline_25K_gpu1.log"
|
|
@@ -39,7 +39,7 @@
|
|
| 39 |
"gpu": 2,
|
| 40 |
"status": "done",
|
| 41 |
"elapsed": 4753,
|
| 42 |
-
"idle_time":
|
| 43 |
"exit_code": 0,
|
| 44 |
"retries": 0,
|
| 45 |
"log_file": "/tmp/gpu_queue/job_002_as_sorl_abs10_K1_25K_gpu2.log"
|
|
@@ -51,7 +51,7 @@
|
|
| 51 |
"gpu": 0,
|
| 52 |
"status": "done",
|
| 53 |
"elapsed": 2366,
|
| 54 |
-
"idle_time":
|
| 55 |
"exit_code": 0,
|
| 56 |
"retries": 0,
|
| 57 |
"log_file": "/tmp/gpu_queue/job_003_add_sub_baseline_50K_gpu0.log"
|
|
@@ -63,7 +63,7 @@
|
|
| 63 |
"gpu": 1,
|
| 64 |
"status": "done",
|
| 65 |
"elapsed": 6727,
|
| 66 |
-
"idle_time":
|
| 67 |
"exit_code": 0,
|
| 68 |
"retries": 0,
|
| 69 |
"log_file": "/tmp/gpu_queue/job_004_as_sorl_abs10_K1_50K_gpu1.log"
|
|
@@ -75,7 +75,7 @@
|
|
| 75 |
"gpu": 2,
|
| 76 |
"status": "done",
|
| 77 |
"elapsed": 3112,
|
| 78 |
-
"idle_time":
|
| 79 |
"exit_code": 0,
|
| 80 |
"retries": 0,
|
| 81 |
"log_file": "/tmp/gpu_queue/job_005_add_sub_baseline_100K_gpu2.log"
|
|
@@ -87,7 +87,7 @@
|
|
| 87 |
"gpu": 0,
|
| 88 |
"status": "done",
|
| 89 |
"elapsed": 8856,
|
| 90 |
-
"idle_time":
|
| 91 |
"exit_code": 0,
|
| 92 |
"retries": 0,
|
| 93 |
"log_file": "/tmp/gpu_queue/job_006_as_sorl_abs10_K1_100K_gpu0.log"
|
|
@@ -459,7 +459,7 @@
|
|
| 459 |
"gpu": 0,
|
| 460 |
"status": "done",
|
| 461 |
"elapsed": 4510,
|
| 462 |
-
"idle_time":
|
| 463 |
"exit_code": 0,
|
| 464 |
"retries": 0,
|
| 465 |
"log_file": "/tmp/gpu_queue/job_037_as_sorl_abs10_K4_25K_gpu0.log"
|
|
@@ -531,7 +531,7 @@
|
|
| 531 |
"gpu": 0,
|
| 532 |
"status": "done",
|
| 533 |
"elapsed": 5510,
|
| 534 |
-
"idle_time":
|
| 535 |
"exit_code": 0,
|
| 536 |
"retries": 0,
|
| 537 |
"log_file": "/tmp/gpu_queue/job_043_as_sorl_abs50_K4_50K_gpu0.log"
|
|
@@ -651,7 +651,7 @@
|
|
| 651 |
"gpu": 1,
|
| 652 |
"status": "failed",
|
| 653 |
"elapsed": 5,
|
| 654 |
-
"idle_time":
|
| 655 |
"exit_code": -9,
|
| 656 |
"retries": 1,
|
| 657 |
"log_file": "/tmp/gpu_queue/job_053_as_sorl_abs10_K4_zipf10.0_500K_gpu1.log"
|
|
@@ -770,8 +770,8 @@
|
|
| 770 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 10 --K 1 --num",
|
| 771 |
"gpu": 0,
|
| 772 |
"status": "running",
|
| 773 |
-
"elapsed":
|
| 774 |
-
"idle_time":
|
| 775 |
"exit_code": -1,
|
| 776 |
"retries": 0,
|
| 777 |
"log_file": "/tmp/gpu_queue/job_063_as_sorl_abs10_K1_50K_1L3H510d_gpu0.log"
|
|
@@ -842,8 +842,8 @@
|
|
| 842 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
|
| 843 |
"gpu": 2,
|
| 844 |
"status": "running",
|
| 845 |
-
"elapsed":
|
| 846 |
-
"idle_time":
|
| 847 |
"exit_code": -1,
|
| 848 |
"retries": 0,
|
| 849 |
"log_file": "/tmp/gpu_queue/job_069_as_sorl_abs10_K1_500K_1L3H510d_gpu2.log"
|
|
@@ -903,7 +903,7 @@
|
|
| 903 |
"gpu": 2,
|
| 904 |
"status": "done",
|
| 905 |
"elapsed": 2797,
|
| 906 |
-
"idle_time":
|
| 907 |
"exit_code": 0,
|
| 908 |
"retries": 0,
|
| 909 |
"log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
|
|
@@ -951,7 +951,7 @@
|
|
| 951 |
"gpu": 2,
|
| 952 |
"status": "done",
|
| 953 |
"elapsed": 7064,
|
| 954 |
-
"idle_time":
|
| 955 |
"exit_code": 0,
|
| 956 |
"retries": 0,
|
| 957 |
"log_file": "/tmp/gpu_queue/job_078_as_baseline_500K_1L2H256d_gpu2.log"
|
|
@@ -975,7 +975,7 @@
|
|
| 975 |
"gpu": 1,
|
| 976 |
"status": "done",
|
| 977 |
"elapsed": 1971,
|
| 978 |
-
"idle_time":
|
| 979 |
"exit_code": 0,
|
| 980 |
"retries": 0,
|
| 981 |
"log_file": "/tmp/gpu_queue/job_080_as_baseline_25K_2L1H128d_gpu1.log"
|
|
@@ -1058,8 +1058,8 @@
|
|
| 1058 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 1 --nu",
|
| 1059 |
"gpu": 1,
|
| 1060 |
"status": "running",
|
| 1061 |
-
"elapsed":
|
| 1062 |
-
"idle_time":
|
| 1063 |
"exit_code": -1,
|
| 1064 |
"retries": 0,
|
| 1065 |
"log_file": "/tmp/gpu_queue/job_087_as_sorl_abs10_K1_250K_2L1H128d_gpu1.log"
|
|
@@ -1082,8 +1082,8 @@
|
|
| 1082 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
|
| 1083 |
"gpu": 1,
|
| 1084 |
"status": "running",
|
| 1085 |
-
"elapsed":
|
| 1086 |
-
"idle_time":
|
| 1087 |
"exit_code": -1,
|
| 1088 |
"retries": 0,
|
| 1089 |
"log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
|
|
@@ -1118,8 +1118,8 @@
|
|
| 1118 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 30 --K 4 --num",
|
| 1119 |
"gpu": 0,
|
| 1120 |
"status": "running",
|
| 1121 |
-
"elapsed":
|
| 1122 |
-
"idle_time":
|
| 1123 |
"exit_code": -1,
|
| 1124 |
"retries": 0,
|
| 1125 |
"log_file": "/tmp/gpu_queue/job_092_as_sorl_abs30_K4_10K_gpu0.log"
|
|
@@ -1154,8 +1154,8 @@
|
|
| 1154 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 4 --nu",
|
| 1155 |
"gpu": 2,
|
| 1156 |
"status": "running",
|
| 1157 |
-
"elapsed":
|
| 1158 |
-
"idle_time":
|
| 1159 |
"exit_code": -1,
|
| 1160 |
"retries": 0,
|
| 1161 |
"log_file": "/tmp/gpu_queue/job_095_as_sorl_abs10_K4_250K_gpu2.log"
|
|
|
|
| 1 |
{
|
| 2 |
+
"timestamp": "2026-04-12 06:20:29",
|
| 3 |
"total": 98,
|
| 4 |
"pending": 79,
|
| 5 |
"running": 6,
|
|
|
|
| 15 |
"gpu": 0,
|
| 16 |
"status": "done",
|
| 17 |
"elapsed": 1786,
|
| 18 |
+
"idle_time": 13967,
|
| 19 |
"exit_code": 0,
|
| 20 |
"retries": 0,
|
| 21 |
"log_file": "/tmp/gpu_queue/job_000_add_sub_baseline_10K_gpu0.log"
|
|
|
|
| 27 |
"gpu": 1,
|
| 28 |
"status": "done",
|
| 29 |
"elapsed": 2101,
|
| 30 |
+
"idle_time": 13652,
|
| 31 |
"exit_code": 0,
|
| 32 |
"retries": 0,
|
| 33 |
"log_file": "/tmp/gpu_queue/job_001_add_sub_baseline_25K_gpu1.log"
|
|
|
|
| 39 |
"gpu": 2,
|
| 40 |
"status": "done",
|
| 41 |
"elapsed": 4753,
|
| 42 |
+
"idle_time": 11000,
|
| 43 |
"exit_code": 0,
|
| 44 |
"retries": 0,
|
| 45 |
"log_file": "/tmp/gpu_queue/job_002_as_sorl_abs10_K1_25K_gpu2.log"
|
|
|
|
| 51 |
"gpu": 0,
|
| 52 |
"status": "done",
|
| 53 |
"elapsed": 2366,
|
| 54 |
+
"idle_time": 13387,
|
| 55 |
"exit_code": 0,
|
| 56 |
"retries": 0,
|
| 57 |
"log_file": "/tmp/gpu_queue/job_003_add_sub_baseline_50K_gpu0.log"
|
|
|
|
| 63 |
"gpu": 1,
|
| 64 |
"status": "done",
|
| 65 |
"elapsed": 6727,
|
| 66 |
+
"idle_time": 9026,
|
| 67 |
"exit_code": 0,
|
| 68 |
"retries": 0,
|
| 69 |
"log_file": "/tmp/gpu_queue/job_004_as_sorl_abs10_K1_50K_gpu1.log"
|
|
|
|
| 75 |
"gpu": 2,
|
| 76 |
"status": "done",
|
| 77 |
"elapsed": 3112,
|
| 78 |
+
"idle_time": 12640,
|
| 79 |
"exit_code": 0,
|
| 80 |
"retries": 0,
|
| 81 |
"log_file": "/tmp/gpu_queue/job_005_add_sub_baseline_100K_gpu2.log"
|
|
|
|
| 87 |
"gpu": 0,
|
| 88 |
"status": "done",
|
| 89 |
"elapsed": 8856,
|
| 90 |
+
"idle_time": 5107,
|
| 91 |
"exit_code": 0,
|
| 92 |
"retries": 0,
|
| 93 |
"log_file": "/tmp/gpu_queue/job_006_as_sorl_abs10_K1_100K_gpu0.log"
|
|
|
|
| 459 |
"gpu": 0,
|
| 460 |
"status": "done",
|
| 461 |
"elapsed": 4510,
|
| 462 |
+
"idle_time": 8872,
|
| 463 |
"exit_code": 0,
|
| 464 |
"retries": 0,
|
| 465 |
"log_file": "/tmp/gpu_queue/job_037_as_sorl_abs10_K4_25K_gpu0.log"
|
|
|
|
| 531 |
"gpu": 0,
|
| 532 |
"status": "done",
|
| 533 |
"elapsed": 5510,
|
| 534 |
+
"idle_time": 3358,
|
| 535 |
"exit_code": 0,
|
| 536 |
"retries": 0,
|
| 537 |
"log_file": "/tmp/gpu_queue/job_043_as_sorl_abs50_K4_50K_gpu0.log"
|
|
|
|
| 651 |
"gpu": 1,
|
| 652 |
"status": "failed",
|
| 653 |
"elapsed": 5,
|
| 654 |
+
"idle_time": 7036,
|
| 655 |
"exit_code": -9,
|
| 656 |
"retries": 1,
|
| 657 |
"log_file": "/tmp/gpu_queue/job_053_as_sorl_abs10_K4_zipf10.0_500K_gpu1.log"
|
|
|
|
| 770 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 10 --K 1 --num",
|
| 771 |
"gpu": 0,
|
| 772 |
"status": "running",
|
| 773 |
+
"elapsed": 5101,
|
| 774 |
+
"idle_time": 27,
|
| 775 |
"exit_code": -1,
|
| 776 |
"retries": 0,
|
| 777 |
"log_file": "/tmp/gpu_queue/job_063_as_sorl_abs10_K1_50K_1L3H510d_gpu0.log"
|
|
|
|
| 842 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
|
| 843 |
"gpu": 2,
|
| 844 |
"status": "running",
|
| 845 |
+
"elapsed": 8196,
|
| 846 |
+
"idle_time": 0,
|
| 847 |
"exit_code": -1,
|
| 848 |
"retries": 0,
|
| 849 |
"log_file": "/tmp/gpu_queue/job_069_as_sorl_abs10_K1_500K_1L3H510d_gpu2.log"
|
|
|
|
| 903 |
"gpu": 2,
|
| 904 |
"status": "done",
|
| 905 |
"elapsed": 2797,
|
| 906 |
+
"idle_time": 8201,
|
| 907 |
"exit_code": 0,
|
| 908 |
"retries": 0,
|
| 909 |
"log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
|
|
|
|
| 951 |
"gpu": 2,
|
| 952 |
"status": "done",
|
| 953 |
"elapsed": 7064,
|
| 954 |
+
"idle_time": 5574,
|
| 955 |
"exit_code": 0,
|
| 956 |
"retries": 0,
|
| 957 |
"log_file": "/tmp/gpu_queue/job_078_as_baseline_500K_1L2H256d_gpu2.log"
|
|
|
|
| 975 |
"gpu": 1,
|
| 976 |
"status": "done",
|
| 977 |
"elapsed": 1971,
|
| 978 |
+
"idle_time": 7052,
|
| 979 |
"exit_code": 0,
|
| 980 |
"retries": 0,
|
| 981 |
"log_file": "/tmp/gpu_queue/job_080_as_baseline_25K_2L1H128d_gpu1.log"
|
|
|
|
| 1058 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 1 --nu",
|
| 1059 |
"gpu": 1,
|
| 1060 |
"status": "running",
|
| 1061 |
+
"elapsed": 7036,
|
| 1062 |
+
"idle_time": 4,
|
| 1063 |
"exit_code": -1,
|
| 1064 |
"retries": 0,
|
| 1065 |
"log_file": "/tmp/gpu_queue/job_087_as_sorl_abs10_K1_250K_2L1H128d_gpu1.log"
|
|
|
|
| 1082 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
|
| 1083 |
"gpu": 1,
|
| 1084 |
"status": "running",
|
| 1085 |
+
"elapsed": 13646,
|
| 1086 |
+
"idle_time": 142,
|
| 1087 |
"exit_code": -1,
|
| 1088 |
"retries": 0,
|
| 1089 |
"log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
|
|
|
|
| 1118 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 30 --K 4 --num",
|
| 1119 |
"gpu": 0,
|
| 1120 |
"status": "running",
|
| 1121 |
+
"elapsed": 3355,
|
| 1122 |
+
"idle_time": 65,
|
| 1123 |
"exit_code": -1,
|
| 1124 |
"retries": 0,
|
| 1125 |
"log_file": "/tmp/gpu_queue/job_092_as_sorl_abs30_K4_10K_gpu0.log"
|
|
|
|
| 1154 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 4 --nu",
|
| 1155 |
"gpu": 2,
|
| 1156 |
"status": "running",
|
| 1157 |
+
"elapsed": 5571,
|
| 1158 |
+
"idle_time": 3,
|
| 1159 |
"exit_code": -1,
|
| 1160 |
"retries": 0,
|
| 1161 |
"log_file": "/tmp/gpu_queue/job_095_as_sorl_abs10_K4_250K_gpu2.log"
|