amirali1985 commited on
Commit
57fbf0c
·
verified ·
1 Parent(s): 6e35604

Upload queue_status.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. queue_status.json +40 -40
queue_status.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "timestamp": "2026-04-15 11:36:00",
3
  "total": 59,
4
- "pending": 31,
5
- "running": 5,
6
  "done": 23,
7
  "failed": 0,
8
  "stale": 0,
@@ -15,7 +15,7 @@
15
  "gpu": 0,
16
  "status": "done",
17
  "elapsed": 19215,
18
- "idle_time": 30887,
19
  "exit_code": 0,
20
  "retries": 0,
21
  "log_file": "/tmp/gpu_queue/job_000_as_sorl_abs20_K1_100K_2L1H128d_gpu0.log"
@@ -27,7 +27,7 @@
27
  "gpu": 1,
28
  "status": "done",
29
  "elapsed": 17962,
30
- "idle_time": 32141,
31
  "exit_code": 0,
32
  "retries": 0,
33
  "log_file": "/tmp/gpu_queue/job_001_as_sorl_abs20_K1_100K_1L2H256d_gpu1.log"
@@ -39,7 +39,7 @@
39
  "gpu": 2,
40
  "status": "done",
41
  "elapsed": 20071,
42
- "idle_time": 30032,
43
  "exit_code": 0,
44
  "retries": 0,
45
  "log_file": "/tmp/gpu_queue/job_002_as_sorl_abs20_K1_100K_1L3H510d_gpu2.log"
@@ -51,7 +51,7 @@
51
  "gpu": 0,
52
  "status": "done",
53
  "elapsed": 19232,
54
- "idle_time": 30871,
55
  "exit_code": 0,
56
  "retries": 0,
57
  "log_file": "/tmp/gpu_queue/job_003_as_sorl_abs1_K1_100K_2L1H128d_gpu0.log"
@@ -63,7 +63,7 @@
63
  "gpu": 1,
64
  "status": "done",
65
  "elapsed": 17980,
66
- "idle_time": 32123,
67
  "exit_code": 0,
68
  "retries": 0,
69
  "log_file": "/tmp/gpu_queue/job_004_as_sorl_abs1_K1_100K_1L2H256d_gpu1.log"
@@ -75,7 +75,7 @@
75
  "gpu": 2,
76
  "status": "done",
77
  "elapsed": 20071,
78
- "idle_time": 30032,
79
  "exit_code": 0,
80
  "retries": 0,
81
  "log_file": "/tmp/gpu_queue/job_005_as_sorl_abs1_K1_100K_1L3H510d_gpu2.log"
@@ -87,7 +87,7 @@
87
  "gpu": 2,
88
  "status": "done",
89
  "elapsed": 17731,
90
- "idle_time": 7899,
91
  "exit_code": 0,
92
  "retries": 0,
93
  "log_file": "/tmp/gpu_queue/job_006_as_sorl_abs2_K1_100K_2L1H128d_gpu2.log"
@@ -99,7 +99,7 @@
99
  "gpu": 1,
100
  "status": "done",
101
  "elapsed": 14745,
102
- "idle_time": 974,
103
  "exit_code": 0,
104
  "retries": 0,
105
  "log_file": "/tmp/gpu_queue/job_007_as_sorl_abs2_K1_100K_1L2H256d_gpu1.log"
@@ -123,7 +123,7 @@
123
  "gpu": 1,
124
  "status": "done",
125
  "elapsed": 15057,
126
- "idle_time": 17064,
127
  "exit_code": 0,
128
  "retries": 0,
129
  "log_file": "/tmp/gpu_queue/job_009_as_sorl_abs5_K1_100K_2L1H128d_gpu1.log"
@@ -134,7 +134,7 @@
134
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 5 --K 1 --num",
135
  "gpu": 0,
136
  "status": "running",
137
- "elapsed": 672,
138
  "idle_time": 7,
139
  "exit_code": -1,
140
  "retries": 0,
@@ -146,8 +146,8 @@
146
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 5 --K 1 --num",
147
  "gpu": 0,
148
  "status": "running",
149
- "elapsed": 12878,
150
- "idle_time": 25,
151
  "exit_code": -1,
152
  "retries": 0,
153
  "log_file": "/tmp/gpu_queue/job_011_as_sorl_abs5_K1_100K_1L3H510d_gpu0.log"
@@ -158,8 +158,8 @@
158
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 50 --K 1 --nu",
159
  "gpu": 2,
160
  "status": "running",
161
- "elapsed": 13403,
162
- "idle_time": 11,
163
  "exit_code": -1,
164
  "retries": 0,
165
  "log_file": "/tmp/gpu_queue/job_012_as_sorl_abs50_K1_100K_2L1H128d_gpu2.log"
@@ -192,13 +192,13 @@
192
  "job_id": 15,
193
  "name": "as_baseline_10K_2L1H128d",
194
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 10000 --num_epochs 20 --n_la",
195
- "gpu": -1,
196
- "status": "pending",
197
- "elapsed": 0,
198
- "idle_time": 0,
199
  "exit_code": -1,
200
  "retries": 0,
201
- "log_file": ""
202
  },
203
  {
204
  "job_id": 16,
@@ -207,7 +207,7 @@
207
  "gpu": 1,
208
  "status": "done",
209
  "elapsed": 1347,
210
- "idle_time": 21118,
211
  "exit_code": 0,
212
  "retries": 0,
213
  "log_file": "/tmp/gpu_queue/job_016_as_baseline_10K_1L2H256d_gpu1.log"
@@ -219,7 +219,7 @@
219
  "gpu": 1,
220
  "status": "done",
221
  "elapsed": 1337,
222
- "idle_time": 15723,
223
  "exit_code": 0,
224
  "retries": 0,
225
  "log_file": "/tmp/gpu_queue/job_017_as_baseline_10K_1L3H510d_gpu1.log"
@@ -231,7 +231,7 @@
231
  "gpu": 1,
232
  "status": "done",
233
  "elapsed": 6071,
234
- "idle_time": 8019,
235
  "exit_code": 0,
236
  "retries": 0,
237
  "log_file": "/tmp/gpu_queue/job_018_as_sorl_abs30_K1_10K_2L1H128d_gpu1.log"
@@ -267,7 +267,7 @@
267
  "gpu": 1,
268
  "status": "done",
269
  "elapsed": 1517,
270
- "idle_time": 22469,
271
  "exit_code": 0,
272
  "retries": 0,
273
  "log_file": "/tmp/gpu_queue/job_021_as_baseline_25K_2L1H128d_gpu1.log"
@@ -278,8 +278,8 @@
278
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --n_la",
279
  "gpu": 1,
280
  "status": "running",
281
- "elapsed": 972,
282
- "idle_time": 29,
283
  "exit_code": -1,
284
  "retries": 0,
285
  "log_file": "/tmp/gpu_queue/job_022_as_baseline_25K_1L2H256d_gpu1.log"
@@ -303,7 +303,7 @@
303
  "gpu": 2,
304
  "status": "done",
305
  "elapsed": 7897,
306
- "idle_time": 2,
307
  "exit_code": 0,
308
  "retries": 0,
309
  "log_file": "/tmp/gpu_queue/job_024_as_sorl_abs30_K1_25K_2L1H128d_gpu2.log"
@@ -315,7 +315,7 @@
315
  "gpu": 1,
316
  "status": "done",
317
  "elapsed": 7025,
318
- "idle_time": 14090,
319
  "exit_code": 0,
320
  "retries": 0,
321
  "log_file": "/tmp/gpu_queue/job_025_as_sorl_abs30_K1_25K_1L2H256d_gpu1.log"
@@ -327,7 +327,7 @@
327
  "gpu": 1,
328
  "status": "done",
329
  "elapsed": 8147,
330
- "idle_time": 23989,
331
  "exit_code": 0,
332
  "retries": 0,
333
  "log_file": "/tmp/gpu_queue/job_026_as_sorl_abs30_K1_25K_1L3H510d_gpu1.log"
@@ -339,7 +339,7 @@
339
  "gpu": 1,
340
  "status": "done",
341
  "elapsed": 1789,
342
- "idle_time": 6225,
343
  "exit_code": 0,
344
  "retries": 0,
345
  "log_file": "/tmp/gpu_queue/job_027_as_baseline_50K_2L1H128d_gpu1.log"
@@ -363,7 +363,7 @@
363
  "gpu": 2,
364
  "status": "done",
365
  "elapsed": 2052,
366
- "idle_time": 25634,
367
  "exit_code": 0,
368
  "retries": 0,
369
  "log_file": "/tmp/gpu_queue/job_029_as_baseline_50K_1L3H510d_gpu2.log"
@@ -375,7 +375,7 @@
375
  "gpu": 0,
376
  "status": "done",
377
  "elapsed": 11753,
378
- "idle_time": 19133,
379
  "exit_code": 0,
380
  "retries": 0,
381
  "log_file": "/tmp/gpu_queue/job_030_as_sorl_abs30_K1_50K_2L1H128d_gpu0.log"
@@ -398,8 +398,8 @@
398
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 30 --K 1 --num",
399
  "gpu": 1,
400
  "status": "running",
401
- "elapsed": 6222,
402
- "idle_time": 16,
403
  "exit_code": -1,
404
  "retries": 0,
405
  "log_file": "/tmp/gpu_queue/job_032_as_sorl_abs30_K1_50K_1L3H510d_gpu1.log"
@@ -411,7 +411,7 @@
411
  "gpu": 2,
412
  "status": "done",
413
  "elapsed": 2343,
414
- "idle_time": 27688,
415
  "exit_code": 0,
416
  "retries": 0,
417
  "log_file": "/tmp/gpu_queue/job_033_as_baseline_100K_2L1H128d_gpu2.log"
@@ -447,7 +447,7 @@
447
  "gpu": 0,
448
  "status": "done",
449
  "elapsed": 18453,
450
- "idle_time": 677,
451
  "exit_code": 0,
452
  "retries": 0,
453
  "log_file": "/tmp/gpu_queue/job_036_as_sorl_abs30_K1_100K_2L1H128d_gpu0.log"
@@ -459,7 +459,7 @@
459
  "gpu": 2,
460
  "status": "done",
461
  "elapsed": 16622,
462
- "idle_time": 13409,
463
  "exit_code": 0,
464
  "retries": 0,
465
  "log_file": "/tmp/gpu_queue/job_037_as_sorl_abs30_K1_100K_1L2H256d_gpu2.log"
@@ -471,7 +471,7 @@
471
  "gpu": 0,
472
  "status": "done",
473
  "elapsed": 17990,
474
- "idle_time": 12882,
475
  "exit_code": 0,
476
  "retries": 0,
477
  "log_file": "/tmp/gpu_queue/job_038_as_sorl_abs30_K1_100K_1L3H510d_gpu0.log"
 
1
  {
2
+ "timestamp": "2026-04-15 11:36:24",
3
  "total": 59,
4
+ "pending": 30,
5
+ "running": 6,
6
  "done": 23,
7
  "failed": 0,
8
  "stale": 0,
 
15
  "gpu": 0,
16
  "status": "done",
17
  "elapsed": 19215,
18
+ "idle_time": 30911,
19
  "exit_code": 0,
20
  "retries": 0,
21
  "log_file": "/tmp/gpu_queue/job_000_as_sorl_abs20_K1_100K_2L1H128d_gpu0.log"
 
27
  "gpu": 1,
28
  "status": "done",
29
  "elapsed": 17962,
30
+ "idle_time": 32165,
31
  "exit_code": 0,
32
  "retries": 0,
33
  "log_file": "/tmp/gpu_queue/job_001_as_sorl_abs20_K1_100K_1L2H256d_gpu1.log"
 
39
  "gpu": 2,
40
  "status": "done",
41
  "elapsed": 20071,
42
+ "idle_time": 30056,
43
  "exit_code": 0,
44
  "retries": 0,
45
  "log_file": "/tmp/gpu_queue/job_002_as_sorl_abs20_K1_100K_1L3H510d_gpu2.log"
 
51
  "gpu": 0,
52
  "status": "done",
53
  "elapsed": 19232,
54
+ "idle_time": 30895,
55
  "exit_code": 0,
56
  "retries": 0,
57
  "log_file": "/tmp/gpu_queue/job_003_as_sorl_abs1_K1_100K_2L1H128d_gpu0.log"
 
63
  "gpu": 1,
64
  "status": "done",
65
  "elapsed": 17980,
66
+ "idle_time": 32147,
67
  "exit_code": 0,
68
  "retries": 0,
69
  "log_file": "/tmp/gpu_queue/job_004_as_sorl_abs1_K1_100K_1L2H256d_gpu1.log"
 
75
  "gpu": 2,
76
  "status": "done",
77
  "elapsed": 20071,
78
+ "idle_time": 30056,
79
  "exit_code": 0,
80
  "retries": 0,
81
  "log_file": "/tmp/gpu_queue/job_005_as_sorl_abs1_K1_100K_1L3H510d_gpu2.log"
 
87
  "gpu": 2,
88
  "status": "done",
89
  "elapsed": 17731,
90
+ "idle_time": 7923,
91
  "exit_code": 0,
92
  "retries": 0,
93
  "log_file": "/tmp/gpu_queue/job_006_as_sorl_abs2_K1_100K_2L1H128d_gpu2.log"
 
99
  "gpu": 1,
100
  "status": "done",
101
  "elapsed": 14745,
102
+ "idle_time": 998,
103
  "exit_code": 0,
104
  "retries": 0,
105
  "log_file": "/tmp/gpu_queue/job_007_as_sorl_abs2_K1_100K_1L2H256d_gpu1.log"
 
123
  "gpu": 1,
124
  "status": "done",
125
  "elapsed": 15057,
126
+ "idle_time": 17088,
127
  "exit_code": 0,
128
  "retries": 0,
129
  "log_file": "/tmp/gpu_queue/job_009_as_sorl_abs5_K1_100K_2L1H128d_gpu1.log"
 
134
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 5 --K 1 --num",
135
  "gpu": 0,
136
  "status": "running",
137
+ "elapsed": 696,
138
  "idle_time": 7,
139
  "exit_code": -1,
140
  "retries": 0,
 
146
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 5 --K 1 --num",
147
  "gpu": 0,
148
  "status": "running",
149
+ "elapsed": 12902,
150
+ "idle_time": 23,
151
  "exit_code": -1,
152
  "retries": 0,
153
  "log_file": "/tmp/gpu_queue/job_011_as_sorl_abs5_K1_100K_1L3H510d_gpu0.log"
 
158
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 50 --K 1 --nu",
159
  "gpu": 2,
160
  "status": "running",
161
+ "elapsed": 13427,
162
+ "idle_time": 5,
163
  "exit_code": -1,
164
  "retries": 0,
165
  "log_file": "/tmp/gpu_queue/job_012_as_sorl_abs50_K1_100K_2L1H128d_gpu2.log"
 
192
  "job_id": 15,
193
  "name": "as_baseline_10K_2L1H128d",
194
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 10000 --num_epochs 20 --n_la",
195
+ "gpu": 2,
196
+ "status": "running",
197
+ "elapsed": 20,
198
+ "idle_time": 7,
199
  "exit_code": -1,
200
  "retries": 0,
201
+ "log_file": "/tmp/gpu_queue/job_015_as_baseline_10K_2L1H128d_gpu2.log"
202
  },
203
  {
204
  "job_id": 16,
 
207
  "gpu": 1,
208
  "status": "done",
209
  "elapsed": 1347,
210
+ "idle_time": 21142,
211
  "exit_code": 0,
212
  "retries": 0,
213
  "log_file": "/tmp/gpu_queue/job_016_as_baseline_10K_1L2H256d_gpu1.log"
 
219
  "gpu": 1,
220
  "status": "done",
221
  "elapsed": 1337,
222
+ "idle_time": 15746,
223
  "exit_code": 0,
224
  "retries": 0,
225
  "log_file": "/tmp/gpu_queue/job_017_as_baseline_10K_1L3H510d_gpu1.log"
 
231
  "gpu": 1,
232
  "status": "done",
233
  "elapsed": 6071,
234
+ "idle_time": 8042,
235
  "exit_code": 0,
236
  "retries": 0,
237
  "log_file": "/tmp/gpu_queue/job_018_as_sorl_abs30_K1_10K_2L1H128d_gpu1.log"
 
267
  "gpu": 1,
268
  "status": "done",
269
  "elapsed": 1517,
270
+ "idle_time": 22493,
271
  "exit_code": 0,
272
  "retries": 0,
273
  "log_file": "/tmp/gpu_queue/job_021_as_baseline_25K_2L1H128d_gpu1.log"
 
278
  "cmd": "python -m arithmetic.train --mode baseline --ops add_sub --dataset_size 25000 --num_epochs 20 --n_la",
279
  "gpu": 1,
280
  "status": "running",
281
+ "elapsed": 996,
282
+ "idle_time": 53,
283
  "exit_code": -1,
284
  "retries": 0,
285
  "log_file": "/tmp/gpu_queue/job_022_as_baseline_25K_1L2H256d_gpu1.log"
 
303
  "gpu": 2,
304
  "status": "done",
305
  "elapsed": 7897,
306
+ "idle_time": 26,
307
  "exit_code": 0,
308
  "retries": 0,
309
  "log_file": "/tmp/gpu_queue/job_024_as_sorl_abs30_K1_25K_2L1H128d_gpu2.log"
 
315
  "gpu": 1,
316
  "status": "done",
317
  "elapsed": 7025,
318
+ "idle_time": 14114,
319
  "exit_code": 0,
320
  "retries": 0,
321
  "log_file": "/tmp/gpu_queue/job_025_as_sorl_abs30_K1_25K_1L2H256d_gpu1.log"
 
327
  "gpu": 1,
328
  "status": "done",
329
  "elapsed": 8147,
330
+ "idle_time": 24013,
331
  "exit_code": 0,
332
  "retries": 0,
333
  "log_file": "/tmp/gpu_queue/job_026_as_sorl_abs30_K1_25K_1L3H510d_gpu1.log"
 
339
  "gpu": 1,
340
  "status": "done",
341
  "elapsed": 1789,
342
+ "idle_time": 6249,
343
  "exit_code": 0,
344
  "retries": 0,
345
  "log_file": "/tmp/gpu_queue/job_027_as_baseline_50K_2L1H128d_gpu1.log"
 
363
  "gpu": 2,
364
  "status": "done",
365
  "elapsed": 2052,
366
+ "idle_time": 25658,
367
  "exit_code": 0,
368
  "retries": 0,
369
  "log_file": "/tmp/gpu_queue/job_029_as_baseline_50K_1L3H510d_gpu2.log"
 
375
  "gpu": 0,
376
  "status": "done",
377
  "elapsed": 11753,
378
+ "idle_time": 19157,
379
  "exit_code": 0,
380
  "retries": 0,
381
  "log_file": "/tmp/gpu_queue/job_030_as_sorl_abs30_K1_50K_2L1H128d_gpu0.log"
 
398
  "cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 50000 --abs_vocab 30 --K 1 --num",
399
  "gpu": 1,
400
  "status": "running",
401
+ "elapsed": 6246,
402
+ "idle_time": 6,
403
  "exit_code": -1,
404
  "retries": 0,
405
  "log_file": "/tmp/gpu_queue/job_032_as_sorl_abs30_K1_50K_1L3H510d_gpu1.log"
 
411
  "gpu": 2,
412
  "status": "done",
413
  "elapsed": 2343,
414
+ "idle_time": 27712,
415
  "exit_code": 0,
416
  "retries": 0,
417
  "log_file": "/tmp/gpu_queue/job_033_as_baseline_100K_2L1H128d_gpu2.log"
 
447
  "gpu": 0,
448
  "status": "done",
449
  "elapsed": 18453,
450
+ "idle_time": 701,
451
  "exit_code": 0,
452
  "retries": 0,
453
  "log_file": "/tmp/gpu_queue/job_036_as_sorl_abs30_K1_100K_2L1H128d_gpu0.log"
 
459
  "gpu": 2,
460
  "status": "done",
461
  "elapsed": 16622,
462
+ "idle_time": 13433,
463
  "exit_code": 0,
464
  "retries": 0,
465
  "log_file": "/tmp/gpu_queue/job_037_as_sorl_abs30_K1_100K_1L2H256d_gpu2.log"
 
471
  "gpu": 0,
472
  "status": "done",
473
  "elapsed": 17990,
474
+ "idle_time": 12906,
475
  "exit_code": 0,
476
  "retries": 0,
477
  "log_file": "/tmp/gpu_queue/job_038_as_sorl_abs30_K1_100K_1L3H510d_gpu0.log"