arithmetic-sorl / queue_status.json
amirali1985's picture
queue status update
97a1a05 verified
raw
history blame
47.9 kB
{
"timestamp": "2026-04-12 08:11:05",
"total": 120,
"pending": 0,
"running": 0,
"done": 15,
"failed": 105,
"stale": 0,
"retrying": 0,
"jobs": [
{
"job_id": 0,
"name": "add_sub_baseline_10K",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 10000 --num_epochs 20 --push",
"gpu": 0,
"status": "done",
"elapsed": 1786,
"idle_time": 20603,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_000_add_sub_baseline_10K_gpu0.log"
},
{
"job_id": 1,
"name": "add_sub_baseline_25K",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --push",
"gpu": 1,
"status": "done",
"elapsed": 2101,
"idle_time": 20288,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_001_add_sub_baseline_25K_gpu1.log"
},
{
"job_id": 2,
"name": "as_sorl_abs10_K1_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 10 --K 1 --num",
"gpu": 2,
"status": "done",
"elapsed": 4753,
"idle_time": 17636,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_002_as_sorl_abs10_K1_25K_gpu2.log"
},
{
"job_id": 3,
"name": "add_sub_baseline_50K",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 50000 --num_epochs 20 --push",
"gpu": 0,
"status": "done",
"elapsed": 2366,
"idle_time": 20022,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_003_add_sub_baseline_50K_gpu0.log"
},
{
"job_id": 4,
"name": "as_sorl_abs10_K1_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 10 --K 1 --num",
"gpu": 1,
"status": "done",
"elapsed": 6727,
"idle_time": 15662,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_004_as_sorl_abs10_K1_50K_gpu1.log"
},
{
"job_id": 5,
"name": "add_sub_baseline_100K",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 100000 --num_epochs 20 --pus",
"gpu": 2,
"status": "done",
"elapsed": 3112,
"idle_time": 19276,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_005_add_sub_baseline_100K_gpu2.log"
},
{
"job_id": 6,
"name": "as_sorl_abs10_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --nu",
"gpu": 0,
"status": "done",
"elapsed": 8856,
"idle_time": 11743,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_006_as_sorl_abs10_K1_100K_gpu0.log"
},
{
"job_id": 7,
"name": "add_sub_baseline_250K",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 250000 --num_epochs 20 --pus",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 45,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_007_add_sub_baseline_250K_gpu0.log"
},
{
"job_id": 8,
"name": "as_sorl_abs10_K1_250K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 1 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 75,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_008_as_sorl_abs10_K1_250K_gpu0.log"
},
{
"job_id": 9,
"name": "add_sub_baseline_500K",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 500000 --num_epochs 20 --pus",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 110,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_009_add_sub_baseline_500K_gpu1.log"
},
{
"job_id": 10,
"name": "as_sorl_abs10_K1_10K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 10 --K 1 --num",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 159,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_010_as_sorl_abs10_K1_10K_gpu2.log"
},
{
"job_id": 11,
"name": "as_sorl_abs2_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 2 --K 4 --num",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 215,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_011_as_sorl_abs2_K4_500K_gpu2.log"
},
{
"job_id": 12,
"name": "as_sorl_abs5_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 5 --K 4 --num",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 175,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_012_as_sorl_abs5_K4_500K_gpu1.log"
},
{
"job_id": 13,
"name": "as_sorl_abs10_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 4 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 1249,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_013_as_sorl_abs10_K4_500K_gpu0.log"
},
{
"job_id": 14,
"name": "as_sorl_abs16_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 16 --K 4 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 140,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_014_as_sorl_abs16_K4_500K_gpu0.log"
},
{
"job_id": 15,
"name": "as_sorl_abs20_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 20 --K 4 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 110,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_015_as_sorl_abs20_K4_500K_gpu2.log"
},
{
"job_id": 16,
"name": "as_sorl_abs50_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 50 --K 4 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 145,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_016_as_sorl_abs50_K4_500K_gpu1.log"
},
{
"job_id": 17,
"name": "as_sorl_abs70_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 70 --K 4 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 145,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_017_as_sorl_abs70_K4_500K_gpu2.log"
},
{
"job_id": 18,
"name": "as_sorl_abs100_K4_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 4 --n",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 205,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_018_as_sorl_abs100_K4_500K_gpu2.log"
},
{
"job_id": 19,
"name": "as_sorl_abs2_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 2 --K 1 --num",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 90,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_019_as_sorl_abs2_K1_500K_gpu1.log"
},
{
"job_id": 20,
"name": "as_sorl_abs5_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 5 --K 1 --num",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 90,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_020_as_sorl_abs5_K1_500K_gpu1.log"
},
{
"job_id": 21,
"name": "as_sorl_abs10_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 154,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_021_as_sorl_abs10_K1_500K_gpu0.log"
},
{
"job_id": 22,
"name": "as_sorl_abs16_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 16 --K 1 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 90,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_022_as_sorl_abs16_K1_500K_gpu0.log"
},
{
"job_id": 23,
"name": "as_sorl_abs20_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 20 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 75,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_023_as_sorl_abs20_K1_500K_gpu1.log"
},
{
"job_id": 24,
"name": "as_sorl_abs50_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 50 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 4,
"idle_time": 136,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_024_as_sorl_abs50_K1_500K_gpu1.log"
},
{
"job_id": 25,
"name": "as_sorl_abs70_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 70 --K 1 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 145,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_025_as_sorl_abs70_K1_500K_gpu2.log"
},
{
"job_id": 26,
"name": "as_sorl_abs100_K1_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 1 --n",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 120,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_026_as_sorl_abs100_K1_500K_gpu1.log"
},
{
"job_id": 27,
"name": "as_sorl_abs5_K1_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 5 --K 1 --num_",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 220,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_027_as_sorl_abs5_K1_25K_gpu0.log"
},
{
"job_id": 28,
"name": "as_sorl_abs30_K1_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 30 --K 1 --num",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 154,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_028_as_sorl_abs30_K1_25K_gpu1.log"
},
{
"job_id": 29,
"name": "as_sorl_abs50_K1_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 50 --K 1 --num",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 85,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_029_as_sorl_abs50_K1_25K_gpu0.log"
},
{
"job_id": 30,
"name": "as_sorl_abs5_K1_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 5 --K 1 --num_",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 95,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_030_as_sorl_abs5_K1_50K_gpu2.log"
},
{
"job_id": 31,
"name": "as_sorl_abs30_K1_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 30 --K 1 --num",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 200,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_031_as_sorl_abs30_K1_50K_gpu2.log"
},
{
"job_id": 32,
"name": "as_sorl_abs50_K1_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 50 --K 1 --num",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 119,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_032_as_sorl_abs50_K1_50K_gpu0.log"
},
{
"job_id": 33,
"name": "as_sorl_abs5_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 5 --K 1 --num",
"gpu": 2,
"status": "failed",
"elapsed": 7,
"idle_time": 159,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_033_as_sorl_abs5_K1_100K_gpu2.log"
},
{
"job_id": 34,
"name": "as_sorl_abs30_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 30 --K 1 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 184,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_034_as_sorl_abs30_K1_100K_gpu0.log"
},
{
"job_id": 35,
"name": "as_sorl_abs50_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 50 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 200,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_035_as_sorl_abs50_K1_100K_gpu1.log"
},
{
"job_id": 36,
"name": "as_sorl_abs5_K4_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 5 --K 4 --num_",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 95,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_036_as_sorl_abs5_K4_25K_gpu2.log"
},
{
"job_id": 37,
"name": "as_sorl_abs10_K4_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 10 --K 4 --num",
"gpu": 0,
"status": "done",
"elapsed": 4510,
"idle_time": 15508,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_037_as_sorl_abs10_K4_25K_gpu0.log"
},
{
"job_id": 38,
"name": "as_sorl_abs30_K4_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 30 --K 4 --num",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 184,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_038_as_sorl_abs30_K4_25K_gpu1.log"
},
{
"job_id": 39,
"name": "as_sorl_abs50_K4_25K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 50 --K 4 --num",
"gpu": 2,
"status": "failed",
"elapsed": 4,
"idle_time": 135,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_039_as_sorl_abs50_K4_25K_gpu2.log"
},
{
"job_id": 40,
"name": "as_sorl_abs5_K4_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 5 --K 4 --num_",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 70,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_040_as_sorl_abs5_K4_50K_gpu2.log"
},
{
"job_id": 41,
"name": "as_sorl_abs10_K4_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 10 --K 4 --num",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 220,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_041_as_sorl_abs10_K4_50K_gpu2.log"
},
{
"job_id": 42,
"name": "as_sorl_abs30_K4_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 30 --K 4 --num",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 60,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_042_as_sorl_abs30_K4_50K_gpu0.log"
},
{
"job_id": 43,
"name": "as_sorl_abs50_K4_50K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 50 --K 4 --num",
"gpu": 0,
"status": "done",
"elapsed": 5510,
"idle_time": 9994,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_043_as_sorl_abs50_K4_50K_gpu0.log"
},
{
"job_id": 44,
"name": "as_sorl_abs5_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 5 --K 4 --num",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 125,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_044_as_sorl_abs5_K4_100K_gpu2.log"
},
{
"job_id": 45,
"name": "as_sorl_abs10_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 4 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 205,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_045_as_sorl_abs10_K4_100K_gpu0.log"
},
{
"job_id": 46,
"name": "as_sorl_abs30_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 30 --K 4 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 190,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_046_as_sorl_abs30_K4_100K_gpu2.log"
},
{
"job_id": 47,
"name": "as_sorl_abs50_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 50 --K 4 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 169,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_047_as_sorl_abs50_K4_100K_gpu0.log"
},
{
"job_id": 48,
"name": "as_sorl_abs10_K1_zipf2.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --al",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 80,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_048_as_sorl_abs10_K1_zipf2.0_500K_gpu0.log"
},
{
"job_id": 49,
"name": "as_sorl_abs10_K1_zipf5.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --al",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 50,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_049_as_sorl_abs10_K1_zipf5.0_500K_gpu1.log"
},
{
"job_id": 50,
"name": "as_sorl_abs10_K1_zipf10.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --al",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 226,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_050_as_sorl_abs10_K1_zipf10.0_500K_gpu2.log"
},
{
"job_id": 51,
"name": "as_sorl_abs10_K4_zipf2.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 4 --al",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 45,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_051_as_sorl_abs10_K4_zipf2.0_500K_gpu0.log"
},
{
"job_id": 52,
"name": "as_sorl_abs10_K4_zipf5.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 4 --al",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 185,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_052_as_sorl_abs10_K4_zipf5.0_500K_gpu0.log"
},
{
"job_id": 53,
"name": "as_sorl_abs10_K4_zipf10.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 4 --al",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 13672,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_053_as_sorl_abs10_K4_zipf10.0_500K_gpu1.log"
},
{
"job_id": 54,
"name": "as_sorl_abs100_K1_zipf2.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 1 --a",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 169,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_054_as_sorl_abs100_K1_zipf2.0_500K_gpu0.log"
},
{
"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": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 6372,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_055_as_sorl_abs100_K1_zipf5.0_500K_gpu0.log"
},
{
"job_id": 56,
"name": "as_sorl_abs100_K1_zipf10.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 1 --a",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 80,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_056_as_sorl_abs100_K1_zipf10.0_500_gpu2.log"
},
{
"job_id": 57,
"name": "as_sorl_abs100_K4_zipf2.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 4 --a",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 175,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_057_as_sorl_abs100_K4_zipf2.0_500K_gpu2.log"
},
{
"job_id": 58,
"name": "as_sorl_abs100_K4_zipf5.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 4 --a",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 205,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_058_as_sorl_abs100_K4_zipf5.0_500K_gpu1.log"
},
{
"job_id": 59,
"name": "as_sorl_abs100_K4_zipf10.0_500K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 100 --K 4 --a",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 169,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_059_as_sorl_abs100_K4_zipf10.0_500_gpu1.log"
},
{
"job_id": 60,
"name": "as_baseline_25K_1L3H510d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --n_la",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 120,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_060_as_baseline_25K_1L3H510d_gpu2.log"
},
{
"job_id": 61,
"name": "as_sorl_abs10_K1_25K_1L3H510d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 10 --K 1 --num",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 65,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_061_as_sorl_abs10_K1_25K_1L3H510d_gpu2.log"
},
{
"job_id": 62,
"name": "as_baseline_50K_1L3H510d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 50000 --num_epochs 20 --n_la",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 215,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_062_as_baseline_50K_1L3H510d_gpu0.log"
},
{
"job_id": 63,
"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": "done",
"elapsed": 5352,
"idle_time": 6386,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_063_as_sorl_abs10_K1_50K_1L3H510d_gpu0.log"
},
{
"job_id": 64,
"name": "as_baseline_100K_1L3H510d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 100000 --num_epochs 20 --n_l",
"gpu": 1,
"status": "failed",
"elapsed": 4,
"idle_time": 236,
"exit_code": -15,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_064_as_baseline_100K_1L3H510d_gpu1.log"
},
{
"job_id": 65,
"name": "as_sorl_abs10_K1_100K_1L3H510d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 226,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_065_as_sorl_abs10_K1_100K_1L3H510d_gpu0.log"
},
{
"job_id": 66,
"name": "as_baseline_250K_1L3H510d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 250000 --num_epochs 20 --n_l",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 200,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_066_as_baseline_250K_1L3H510d_gpu0.log"
},
{
"job_id": 67,
"name": "as_sorl_abs10_K1_250K_1L3H510d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 125,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_067_as_sorl_abs10_K1_250K_1L3H510d_gpu1.log"
},
{
"job_id": 68,
"name": "as_baseline_500K_1L3H510d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 500000 --num_epochs 20 --n_l",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 105,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_068_as_baseline_500K_1L3H510d_gpu1.log"
},
{
"job_id": 69,
"name": "as_sorl_abs10_K1_500K_1L3H510d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 2,
"idle_time": 234,
"exit_code": -15,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_069_as_sorl_abs10_K1_500K_1L3H510d_gpu2.log"
},
{
"job_id": 70,
"name": "as_baseline_25K_1L2H256d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --n_la",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 226,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_070_as_baseline_25K_1L2H256d_gpu1.log"
},
{
"job_id": 71,
"name": "as_sorl_abs10_K1_25K_1L2H256d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 10 --K 1 --num",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 190,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_071_as_sorl_abs10_K1_25K_1L2H256d_gpu0.log"
},
{
"job_id": 72,
"name": "as_baseline_50K_1L2H256d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 50000 --num_epochs 20 --n_la",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 55,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_072_as_baseline_50K_1L2H256d_gpu2.log"
},
{
"job_id": 73,
"name": "as_sorl_abs10_K1_50K_1L2H256d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 10 --K 1 --num",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 75,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_073_as_sorl_abs10_K1_50K_1L2H256d_gpu1.log"
},
{
"job_id": 74,
"name": "as_baseline_100K_1L2H256d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 100000 --num_epochs 20 --n_l",
"gpu": 2,
"status": "done",
"elapsed": 2797,
"idle_time": 14836,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_074_as_baseline_100K_1L2H256d_gpu2.log"
},
{
"job_id": 75,
"name": "as_sorl_abs10_K1_100K_1L2H256d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 220,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_075_as_sorl_abs10_K1_100K_1L2H256d_gpu1.log"
},
{
"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": 0,
"status": "done",
"elapsed": 4832,
"idle_time": 1682,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_076_as_baseline_250K_1L2H256d_gpu0.log"
},
{
"job_id": 77,
"name": "as_sorl_abs10_K1_250K_1L2H256d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 1 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 49,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_077_as_sorl_abs10_K1_250K_1L2H256d_gpu2.log"
},
{
"job_id": 78,
"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": "done",
"elapsed": 7064,
"idle_time": 12209,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_078_as_baseline_500K_1L2H256d_gpu2.log"
},
{
"job_id": 79,
"name": "as_sorl_abs10_K1_500K_1L2H256d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 4,
"idle_time": 136,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_079_as_sorl_abs10_K1_500K_1L2H256d_gpu2.log"
},
{
"job_id": 80,
"name": "as_baseline_25K_2L1H128d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --n_la",
"gpu": 1,
"status": "done",
"elapsed": 1971,
"idle_time": 13688,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_080_as_baseline_25K_2L1H128d_gpu1.log"
},
{
"job_id": 81,
"name": "as_sorl_abs10_K1_25K_2L1H128d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 25000 --abs_vocab 10 --K 1 --num",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 175,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_081_as_sorl_abs10_K1_25K_2L1H128d_gpu2.log"
},
{
"job_id": 82,
"name": "as_baseline_50K_2L1H128d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 50000 --num_epochs 20 --n_la",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 155,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_082_as_baseline_50K_2L1H128d_gpu0.log"
},
{
"job_id": 83,
"name": "as_sorl_abs10_K1_50K_2L1H128d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 10 --K 1 --num",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 60,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_083_as_sorl_abs10_K1_50K_2L1H128d_gpu0.log"
},
{
"job_id": 84,
"name": "as_baseline_100K_2L1H128d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 100000 --num_epochs 20 --n_l",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 125,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_084_as_baseline_100K_2L1H128d_gpu0.log"
},
{
"job_id": 85,
"name": "as_sorl_abs10_K1_100K_2L1H128d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 105,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_085_as_sorl_abs10_K1_100K_2L1H128d_gpu0.log"
},
{
"job_id": 86,
"name": "as_baseline_250K_2L1H128d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 250000 --num_epochs 20 --n_l",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 140,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_086_as_baseline_250K_2L1H128d_gpu1.log"
},
{
"job_id": 87,
"name": "as_sorl_abs10_K1_250K_2L1H128d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 10 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 241,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_087_as_sorl_abs10_K1_250K_2L1H128d_gpu1.log"
},
{
"job_id": 88,
"name": "as_baseline_500K_2L1H128d",
"cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 500000 --num_epochs 20 --n_l",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 60,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_088_as_baseline_500K_2L1H128d_gpu1.log"
},
{
"job_id": 89,
"name": "as_sorl_abs10_K1_500K_2L1H128d",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 500000 --abs_vocab 10 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 241,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_089_as_sorl_abs10_K1_500K_2L1H128d_gpu1.log"
},
{
"job_id": 90,
"name": "as_sorl_abs5_K4_10K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 5 --K 4 --num_",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 65,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_090_as_sorl_abs5_K4_10K_gpu0.log"
},
{
"job_id": 91,
"name": "as_sorl_abs10_K4_10K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 10 --K 4 --num",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 140,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_091_as_sorl_abs10_K4_10K_gpu0.log"
},
{
"job_id": 92,
"name": "as_sorl_abs30_K4_10K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 30 --K 4 --num",
"gpu": 0,
"status": "done",
"elapsed": 3477,
"idle_time": 6517,
"exit_code": 0,
"retries": 0,
"log_file": "/tmp/gpu_queue/job_092_as_sorl_abs30_K4_10K_gpu0.log"
},
{
"job_id": 93,
"name": "as_sorl_abs50_K4_10K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 50 --K 4 --num",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 159,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_093_as_sorl_abs50_K4_10K_gpu1.log"
},
{
"job_id": 94,
"name": "as_sorl_abs5_K4_250K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 5 --K 4 --num",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 100,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_094_as_sorl_abs5_K4_250K_gpu0.log"
},
{
"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": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 237,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_095_as_sorl_abs10_K4_250K_gpu2.log"
},
{
"job_id": 96,
"name": "as_sorl_abs30_K4_250K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 30 --K 4 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 104,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_096_as_sorl_abs30_K4_250K_gpu1.log"
},
{
"job_id": 97,
"name": "as_sorl_abs50_K4_250K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 250000 --abs_vocab 50 --K 4 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 190,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_097_as_sorl_abs50_K4_250K_gpu1.log"
},
{
"job_id": 98,
"name": "as_sorl_abs2_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 2 --K 1 --num",
"gpu": 1,
"status": "failed",
"elapsed": 3,
"idle_time": 235,
"exit_code": -15,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_098_as_sorl_abs2_K1_100K_gpu1.log"
},
{
"job_id": 99,
"name": "as_sorl_abs2_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 2 --K 4 --num",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 215,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_099_as_sorl_abs2_K4_100K_gpu1.log"
},
{
"job_id": 100,
"name": "as_sorl_abs16_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 16 --K 1 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 114,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_100_as_sorl_abs16_K1_100K_gpu0.log"
},
{
"job_id": 101,
"name": "as_sorl_abs16_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 16 --K 4 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 0,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_101_as_sorl_abs16_K4_100K_gpu1.log"
},
{
"job_id": 102,
"name": "as_sorl_abs20_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 20 --K 1 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 40,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_102_as_sorl_abs20_K1_100K_gpu2.log"
},
{
"job_id": 103,
"name": "as_sorl_abs20_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 20 --K 4 --nu",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 0,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_103_as_sorl_abs20_K4_100K_gpu0.log"
},
{
"job_id": 104,
"name": "as_sorl_abs70_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 70 --K 1 --nu",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 30,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_104_as_sorl_abs70_K1_100K_gpu1.log"
},
{
"job_id": 105,
"name": "as_sorl_abs70_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 70 --K 4 --nu",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 35,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_105_as_sorl_abs70_K4_100K_gpu2.log"
},
{
"job_id": 106,
"name": "as_sorl_abs100_K1_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 1 --n",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 15,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_106_as_sorl_abs100_K1_100K_gpu1.log"
},
{
"job_id": 107,
"name": "as_sorl_abs100_K4_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 4 --n",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 24,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_107_as_sorl_abs100_K4_100K_gpu0.log"
},
{
"job_id": 108,
"name": "as_sorl_abs10_K1_zipf2.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --al",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 10,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_108_as_sorl_abs10_K1_zipf2.0_100K_gpu0.log"
},
{
"job_id": 109,
"name": "as_sorl_abs10_K4_zipf2.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 4 --al",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 30,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_109_as_sorl_abs10_K4_zipf2.0_100K_gpu1.log"
},
{
"job_id": 110,
"name": "as_sorl_abs10_K1_zipf5.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --al",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 24,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_110_as_sorl_abs10_K1_zipf5.0_100K_gpu1.log"
},
{
"job_id": 111,
"name": "as_sorl_abs10_K4_zipf5.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 4 --al",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 5,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_111_as_sorl_abs10_K4_zipf5.0_100K_gpu2.log"
},
{
"job_id": 112,
"name": "as_sorl_abs10_K1_zipf10.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 1 --al",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 14,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_112_as_sorl_abs10_K1_zipf10.0_100K_gpu2.log"
},
{
"job_id": 113,
"name": "as_sorl_abs10_K4_zipf10.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 10 --K 4 --al",
"gpu": 1,
"status": "failed",
"elapsed": 6,
"idle_time": 10,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_113_as_sorl_abs10_K4_zipf10.0_100K_gpu1.log"
},
{
"job_id": 114,
"name": "as_sorl_abs100_K1_zipf2.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 1 --a",
"gpu": 2,
"status": "failed",
"elapsed": 6,
"idle_time": 19,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_114_as_sorl_abs100_K1_zipf2.0_100K_gpu2.log"
},
{
"job_id": 115,
"name": "as_sorl_abs100_K4_zipf2.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 4 --a",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 14,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_115_as_sorl_abs100_K4_zipf2.0_100K_gpu0.log"
},
{
"job_id": 116,
"name": "as_sorl_abs100_K1_zipf5.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 1 --a",
"gpu": 2,
"status": "failed",
"elapsed": 5,
"idle_time": 0,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_116_as_sorl_abs100_K1_zipf5.0_100K_gpu2.log"
},
{
"job_id": 117,
"name": "as_sorl_abs100_K4_zipf5.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 4 --a",
"gpu": 1,
"status": "failed",
"elapsed": 5,
"idle_time": 44,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_117_as_sorl_abs100_K4_zipf5.0_100K_gpu1.log"
},
{
"job_id": 118,
"name": "as_sorl_abs100_K1_zipf10.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 1 --a",
"gpu": 0,
"status": "failed",
"elapsed": 5,
"idle_time": 36,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_118_as_sorl_abs100_K1_zipf10.0_100_gpu0.log"
},
{
"job_id": 119,
"name": "as_sorl_abs100_K4_zipf10.0_100K",
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 100 --K 4 --a",
"gpu": 0,
"status": "failed",
"elapsed": 6,
"idle_time": 30,
"exit_code": -9,
"retries": 1,
"log_file": "/tmp/gpu_queue/job_119_as_sorl_abs100_K4_zipf10.0_100_gpu0.log"
}
]
}