amirali1985 commited on
Commit
50bbf87
·
verified ·
1 Parent(s): 1da7558

queue status update

Browse files
Files changed (1) hide show
  1. queue_status.json +19 -19
queue_status.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "timestamp": "2026-04-12 03:59:59",
3
  "total": 98,
4
  "pending": 85,
5
  "running": 6,
@@ -15,7 +15,7 @@
15
  "gpu": 0,
16
  "status": "done",
17
  "elapsed": 1786,
18
- "idle_time": 5537,
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": 5222,
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": 2570,
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": 4956,
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": 596,
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": 4210,
79
  "exit_code": 0,
80
  "retries": 0,
81
  "log_file": "/tmp/gpu_queue/job_005_add_sub_baseline_100K_gpu2.log"
@@ -86,8 +86,8 @@
86
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --nu",
87
  "gpu": 0,
88
  "status": "running",
89
- "elapsed": 5530,
90
- "idle_time": 8,
91
  "exit_code": -1,
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": 442,
463
  "exit_code": 0,
464
  "retries": 0,
465
  "log_file": "/tmp/gpu_queue/job_037_as_sorl_abs10_K4_25K_gpu0.log"
@@ -530,8 +530,8 @@
530
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 50 --K 4 --num",
531
  "gpu": 0,
532
  "status": "running",
533
- "elapsed": 435,
534
- "idle_time": 166,
535
  "exit_code": -1,
536
  "retries": 0,
537
  "log_file": "/tmp/gpu_queue/job_043_as_sorl_abs50_K4_50K_gpu0.log"
@@ -902,8 +902,8 @@
902
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 100000 --num_epochs 20 --n_l",
903
  "gpu": 2,
904
  "status": "running",
905
- "elapsed": 2565,
906
- "idle_time": 71,
907
  "exit_code": -1,
908
  "retries": 0,
909
  "log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
@@ -950,7 +950,7 @@
950
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 500000 --num_epochs 20 --n_l",
951
  "gpu": 2,
952
  "status": "running",
953
- "elapsed": 4205,
954
  "idle_time": 1,
955
  "exit_code": -1,
956
  "retries": 0,
@@ -974,8 +974,8 @@
974
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --n_la",
975
  "gpu": 1,
976
  "status": "running",
977
- "elapsed": 590,
978
- "idle_time": 1,
979
  "exit_code": -1,
980
  "retries": 0,
981
  "log_file": "/tmp/gpu_queue/job_080_as_baseline_25K_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": 5215,
1086
- "idle_time": 18,
1087
  "exit_code": -1,
1088
  "retries": 0,
1089
  "log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
 
1
  {
2
+ "timestamp": "2026-04-12 04:01:59",
3
  "total": 98,
4
  "pending": 85,
5
  "running": 6,
 
15
  "gpu": 0,
16
  "status": "done",
17
  "elapsed": 1786,
18
+ "idle_time": 5657,
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": 5342,
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": 2690,
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": 5076,
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": 716,
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": 4330,
79
  "exit_code": 0,
80
  "retries": 0,
81
  "log_file": "/tmp/gpu_queue/job_005_add_sub_baseline_100K_gpu2.log"
 
86
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --nu",
87
  "gpu": 0,
88
  "status": "running",
89
+ "elapsed": 5650,
90
+ "idle_time": 76,
91
  "exit_code": -1,
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": 562,
463
  "exit_code": 0,
464
  "retries": 0,
465
  "log_file": "/tmp/gpu_queue/job_037_as_sorl_abs10_K4_25K_gpu0.log"
 
530
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 50 --K 4 --num",
531
  "gpu": 0,
532
  "status": "running",
533
+ "elapsed": 555,
534
+ "idle_time": 286,
535
  "exit_code": -1,
536
  "retries": 0,
537
  "log_file": "/tmp/gpu_queue/job_043_as_sorl_abs50_K4_50K_gpu0.log"
 
902
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 100000 --num_epochs 20 --n_l",
903
  "gpu": 2,
904
  "status": "running",
905
+ "elapsed": 2685,
906
+ "idle_time": 54,
907
  "exit_code": -1,
908
  "retries": 0,
909
  "log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
 
950
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 500000 --num_epochs 20 --n_l",
951
  "gpu": 2,
952
  "status": "running",
953
+ "elapsed": 4325,
954
  "idle_time": 1,
955
  "exit_code": -1,
956
  "retries": 0,
 
974
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --n_la",
975
  "gpu": 1,
976
  "status": "running",
977
+ "elapsed": 710,
978
+ "idle_time": 24,
979
  "exit_code": -1,
980
  "retries": 0,
981
  "log_file": "/tmp/gpu_queue/job_080_as_baseline_25K_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": 5335,
1086
+ "idle_time": 5,
1087
  "exit_code": -1,
1088
  "retries": 0,
1089
  "log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"