amirali1985 commited on
Commit
47a3e6e
·
verified ·
1 Parent(s): d43d5ff

queue status update

Browse files
Files changed (1) hide show
  1. queue_status.json +34 -34
queue_status.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "timestamp": "2026-04-12 06:24:53",
3
  "total": 98,
4
- "pending": 77,
5
- "running": 5,
6
  "done": 14,
7
  "failed": 2,
8
  "stale": 0,
@@ -15,7 +15,7 @@
15
  "gpu": 0,
16
  "status": "done",
17
  "elapsed": 1786,
18
- "idle_time": 14231,
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": 13916,
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": 11264,
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": 13650,
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": 9290,
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": 12904,
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": 5371,
91
  "exit_code": 0,
92
  "retries": 0,
93
  "log_file": "/tmp/gpu_queue/job_006_as_sorl_abs10_K1_100K_gpu0.log"
@@ -168,13 +168,13 @@
168
  "job_id": 13,
169
  "name": "as_sorl_abs10_K4_500K",
170
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 4 --nu",
171
- "gpu": -1,
172
- "status": "pending",
173
- "elapsed": 0,
174
- "idle_time": 0,
175
  "exit_code": -1,
176
  "retries": 0,
177
- "log_file": ""
178
  },
179
  {
180
  "job_id": 14,
@@ -459,7 +459,7 @@
459
  "gpu": 0,
460
  "status": "done",
461
  "elapsed": 4510,
462
- "idle_time": 9136,
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": 3622,
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": 7300,
655
  "exit_code": -9,
656
  "retries": 1,
657
  "log_file": "/tmp/gpu_queue/job_053_as_sorl_abs10_K4_zipf10.0_500K_gpu1.log"
@@ -675,7 +675,7 @@
675
  "gpu": 0,
676
  "status": "failed",
677
  "elapsed": 5,
678
- "idle_time": 0,
679
  "exit_code": -9,
680
  "retries": 1,
681
  "log_file": "/tmp/gpu_queue/job_055_as_sorl_abs100_K1_zipf5.0_500K_gpu0.log"
@@ -771,7 +771,7 @@
771
  "gpu": 0,
772
  "status": "done",
773
  "elapsed": 5352,
774
- "idle_time": 14,
775
  "exit_code": 0,
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": 8459,
846
- "idle_time": 12,
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": 8464,
907
  "exit_code": 0,
908
  "retries": 0,
909
  "log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
@@ -926,8 +926,8 @@
926
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 250000 --num_epochs 20 --n_l",
927
  "gpu": 0,
928
  "status": "running",
929
- "elapsed": 139,
930
- "idle_time": 27,
931
  "exit_code": -1,
932
  "retries": 0,
933
  "log_file": "/tmp/gpu_queue/job_076_as_baseline_250K_1L2H256d_gpu0.log"
@@ -951,7 +951,7 @@
951
  "gpu": 2,
952
  "status": "done",
953
  "elapsed": 7064,
954
- "idle_time": 5837,
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": 7316,
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": 7299,
1062
- "idle_time": 7,
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": 13909,
1086
- "idle_time": 13,
1087
  "exit_code": -1,
1088
  "retries": 0,
1089
  "log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
@@ -1119,7 +1119,7 @@
1119
  "gpu": 0,
1120
  "status": "done",
1121
  "elapsed": 3477,
1122
- "idle_time": 145,
1123
  "exit_code": 0,
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": 5834,
1158
- "idle_time": 8,
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 06:26:29",
3
  "total": 98,
4
+ "pending": 76,
5
+ "running": 6,
6
  "done": 14,
7
  "failed": 2,
8
  "stale": 0,
 
15
  "gpu": 0,
16
  "status": "done",
17
  "elapsed": 1786,
18
+ "idle_time": 14327,
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": 14012,
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": 11360,
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": 13747,
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": 9386,
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": 13000,
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": 5467,
91
  "exit_code": 0,
92
  "retries": 0,
93
  "log_file": "/tmp/gpu_queue/job_006_as_sorl_abs10_K1_100K_gpu0.log"
 
168
  "job_id": 13,
169
  "name": "as_sorl_abs10_K4_500K",
170
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 4 --nu",
171
+ "gpu": 0,
172
+ "status": "running",
173
+ "elapsed": 95,
174
+ "idle_time": 8,
175
  "exit_code": -1,
176
  "retries": 0,
177
+ "log_file": "/tmp/gpu_queue/job_013_as_sorl_abs10_K4_500K_gpu0.log"
178
  },
179
  {
180
  "job_id": 14,
 
459
  "gpu": 0,
460
  "status": "done",
461
  "elapsed": 4510,
462
+ "idle_time": 9232,
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": 3718,
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": 7396,
655
  "exit_code": -9,
656
  "retries": 1,
657
  "log_file": "/tmp/gpu_queue/job_053_as_sorl_abs10_K4_zipf10.0_500K_gpu1.log"
 
675
  "gpu": 0,
676
  "status": "failed",
677
  "elapsed": 5,
678
+ "idle_time": 96,
679
  "exit_code": -9,
680
  "retries": 1,
681
  "log_file": "/tmp/gpu_queue/job_055_as_sorl_abs100_K1_zipf5.0_500K_gpu0.log"
 
771
  "gpu": 0,
772
  "status": "done",
773
  "elapsed": 5352,
774
+ "idle_time": 110,
775
  "exit_code": 0,
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": 8556,
846
+ "idle_time": 7,
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": 8561,
907
  "exit_code": 0,
908
  "retries": 0,
909
  "log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
 
926
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 250000 --num_epochs 20 --n_l",
927
  "gpu": 0,
928
  "status": "running",
929
+ "elapsed": 235,
930
+ "idle_time": 1,
931
  "exit_code": -1,
932
  "retries": 0,
933
  "log_file": "/tmp/gpu_queue/job_076_as_baseline_250K_1L2H256d_gpu0.log"
 
951
  "gpu": 2,
952
  "status": "done",
953
  "elapsed": 7064,
954
+ "idle_time": 5934,
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": 7412,
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": 7396,
1062
+ "idle_time": 2,
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": 14006,
1086
+ "idle_time": 9,
1087
  "exit_code": -1,
1088
  "retries": 0,
1089
  "log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
 
1119
  "gpu": 0,
1120
  "status": "done",
1121
  "elapsed": 3477,
1122
+ "idle_time": 241,
1123
  "exit_code": 0,
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": 5931,
1158
+ "idle_time": 1,
1159
  "exit_code": -1,
1160
  "retries": 0,
1161
  "log_file": "/tmp/gpu_queue/job_095_as_sorl_abs10_K4_250K_gpu2.log"