Upload queue_status.json with huggingface_hub
Browse files- queue_status.json +27 -27
queue_status.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
{
|
| 2 |
-
"timestamp": "2026-04-16 05:35:
|
| 3 |
"total": 20,
|
| 4 |
"pending": 4,
|
| 5 |
-
"running":
|
| 6 |
-
"done":
|
| 7 |
"failed": 1,
|
| 8 |
"stale": 0,
|
| 9 |
"retrying": 0,
|
|
@@ -15,7 +15,7 @@
|
|
| 15 |
"gpu": 0,
|
| 16 |
"status": "failed",
|
| 17 |
"elapsed": 7,
|
| 18 |
-
"idle_time":
|
| 19 |
"exit_code": 2,
|
| 20 |
"retries": 1,
|
| 21 |
"log_file": "/tmp/gpu_queue/job_000_as_sorl_abs30_K1_100K_emlr5_2L_gpu0.log"
|
|
@@ -27,7 +27,7 @@
|
|
| 27 |
"gpu": 1,
|
| 28 |
"status": "done",
|
| 29 |
"elapsed": 10909,
|
| 30 |
-
"idle_time":
|
| 31 |
"exit_code": 0,
|
| 32 |
"retries": 0,
|
| 33 |
"log_file": "/tmp/gpu_queue/job_001_as_sorl_abs30_K1_50K_ig1_2L1H1_gpu1.log"
|
|
@@ -39,7 +39,7 @@
|
|
| 39 |
"gpu": 2,
|
| 40 |
"status": "done",
|
| 41 |
"elapsed": 10932,
|
| 42 |
-
"idle_time":
|
| 43 |
"exit_code": 0,
|
| 44 |
"retries": 0,
|
| 45 |
"log_file": "/tmp/gpu_queue/job_002_as_sorl_abs30_K1_50K_ig1_1L3H5_gpu2.log"
|
|
@@ -51,7 +51,7 @@
|
|
| 51 |
"gpu": 0,
|
| 52 |
"status": "done",
|
| 53 |
"elapsed": 9082,
|
| 54 |
-
"idle_time":
|
| 55 |
"exit_code": 0,
|
| 56 |
"retries": 0,
|
| 57 |
"log_file": "/tmp/gpu_queue/job_003_as_sorl_abs30_K1_50K_ig5_2L1H1_gpu0.log"
|
|
@@ -63,7 +63,7 @@
|
|
| 63 |
"gpu": 1,
|
| 64 |
"status": "done",
|
| 65 |
"elapsed": 10997,
|
| 66 |
-
"idle_time":
|
| 67 |
"exit_code": 0,
|
| 68 |
"retries": 0,
|
| 69 |
"log_file": "/tmp/gpu_queue/job_004_as_sorl_abs30_K1_50K_ig5_1L3H5_gpu1.log"
|
|
@@ -75,7 +75,7 @@
|
|
| 75 |
"gpu": 2,
|
| 76 |
"status": "done",
|
| 77 |
"elapsed": 10836,
|
| 78 |
-
"idle_time":
|
| 79 |
"exit_code": 0,
|
| 80 |
"retries": 0,
|
| 81 |
"log_file": "/tmp/gpu_queue/job_005_as_sorl_abs30_K1_50K_abs05_2L1_gpu2.log"
|
|
@@ -122,8 +122,8 @@
|
|
| 122 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 30 --K 4 --nu",
|
| 123 |
"gpu": 0,
|
| 124 |
"status": "running",
|
| 125 |
-
"elapsed":
|
| 126 |
-
"idle_time":
|
| 127 |
"exit_code": -1,
|
| 128 |
"retries": 0,
|
| 129 |
"log_file": "/tmp/gpu_queue/job_009_as_sorl_abs30_K4_100K_2L1H128d_gpu0.log"
|
|
@@ -146,8 +146,8 @@
|
|
| 146 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 30 --K 4 --nu",
|
| 147 |
"gpu": 0,
|
| 148 |
"status": "running",
|
| 149 |
-
"elapsed":
|
| 150 |
-
"idle_time":
|
| 151 |
"exit_code": -1,
|
| 152 |
"retries": 0,
|
| 153 |
"log_file": "/tmp/gpu_queue/job_011_as_sorl_abs30_K4_100K_1L3H510d_gpu0.log"
|
|
@@ -159,7 +159,7 @@
|
|
| 159 |
"gpu": 0,
|
| 160 |
"status": "done",
|
| 161 |
"elapsed": 1450,
|
| 162 |
-
"idle_time":
|
| 163 |
"exit_code": 0,
|
| 164 |
"retries": 0,
|
| 165 |
"log_file": "/tmp/gpu_queue/job_012_as_baseline_10K_2L3H510d_gpu0.log"
|
|
@@ -170,8 +170,8 @@
|
|
| 170 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 1 --K 1 --num_",
|
| 171 |
"gpu": 2,
|
| 172 |
"status": "running",
|
| 173 |
-
"elapsed":
|
| 174 |
-
"idle_time":
|
| 175 |
"exit_code": -1,
|
| 176 |
"retries": 0,
|
| 177 |
"log_file": "/tmp/gpu_queue/job_013_as_sorl_abs1_K1_10K_2L3H510d_gpu2.log"
|
|
@@ -182,8 +182,8 @@
|
|
| 182 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 2 --K 1 --num_",
|
| 183 |
"gpu": 1,
|
| 184 |
"status": "running",
|
| 185 |
-
"elapsed":
|
| 186 |
-
"idle_time":
|
| 187 |
"exit_code": -1,
|
| 188 |
"retries": 0,
|
| 189 |
"log_file": "/tmp/gpu_queue/job_014_as_sorl_abs2_K1_10K_2L3H510d_gpu1.log"
|
|
@@ -195,7 +195,7 @@
|
|
| 195 |
"gpu": 0,
|
| 196 |
"status": "done",
|
| 197 |
"elapsed": 5222,
|
| 198 |
-
"idle_time":
|
| 199 |
"exit_code": 0,
|
| 200 |
"retries": 0,
|
| 201 |
"log_file": "/tmp/gpu_queue/job_015_as_sorl_abs5_K1_10K_2L3H510d_gpu0.log"
|
|
@@ -206,8 +206,8 @@
|
|
| 206 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 10 --K 1 --num",
|
| 207 |
"gpu": 1,
|
| 208 |
"status": "running",
|
| 209 |
-
"elapsed":
|
| 210 |
-
"idle_time":
|
| 211 |
"exit_code": -1,
|
| 212 |
"retries": 0,
|
| 213 |
"log_file": "/tmp/gpu_queue/job_016_as_sorl_abs10_K1_10K_2L3H510d_gpu1.log"
|
|
@@ -217,10 +217,10 @@
|
|
| 217 |
"name": "as_sorl_abs20_K1_10K_2L3H510d",
|
| 218 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 20 --K 1 --num",
|
| 219 |
"gpu": 2,
|
| 220 |
-
"status": "
|
| 221 |
-
"elapsed":
|
| 222 |
-
"idle_time":
|
| 223 |
-
"exit_code":
|
| 224 |
"retries": 0,
|
| 225 |
"log_file": "/tmp/gpu_queue/job_017_as_sorl_abs20_K1_10K_2L3H510d_gpu2.log"
|
| 226 |
},
|
|
@@ -231,7 +231,7 @@
|
|
| 231 |
"gpu": 0,
|
| 232 |
"status": "done",
|
| 233 |
"elapsed": 5528,
|
| 234 |
-
"idle_time":
|
| 235 |
"exit_code": 0,
|
| 236 |
"retries": 0,
|
| 237 |
"log_file": "/tmp/gpu_queue/job_018_as_sorl_abs30_K1_10K_2L3H510d_gpu0.log"
|
|
@@ -243,7 +243,7 @@
|
|
| 243 |
"gpu": 0,
|
| 244 |
"status": "done",
|
| 245 |
"elapsed": 5747,
|
| 246 |
-
"idle_time":
|
| 247 |
"exit_code": 0,
|
| 248 |
"retries": 0,
|
| 249 |
"log_file": "/tmp/gpu_queue/job_019_as_sorl_abs50_K1_10K_2L3H510d_gpu0.log"
|
|
|
|
| 1 |
{
|
| 2 |
+
"timestamp": "2026-04-16 05:35:27",
|
| 3 |
"total": 20,
|
| 4 |
"pending": 4,
|
| 5 |
+
"running": 5,
|
| 6 |
+
"done": 10,
|
| 7 |
"failed": 1,
|
| 8 |
"stale": 0,
|
| 9 |
"retrying": 0,
|
|
|
|
| 15 |
"gpu": 0,
|
| 16 |
"status": "failed",
|
| 17 |
"elapsed": 7,
|
| 18 |
+
"idle_time": 15909,
|
| 19 |
"exit_code": 2,
|
| 20 |
"retries": 1,
|
| 21 |
"log_file": "/tmp/gpu_queue/job_000_as_sorl_abs30_K1_100K_emlr5_2L_gpu0.log"
|
|
|
|
| 27 |
"gpu": 1,
|
| 28 |
"status": "done",
|
| 29 |
"elapsed": 10909,
|
| 30 |
+
"idle_time": 5018,
|
| 31 |
"exit_code": 0,
|
| 32 |
"retries": 0,
|
| 33 |
"log_file": "/tmp/gpu_queue/job_001_as_sorl_abs30_K1_50K_ig1_2L1H1_gpu1.log"
|
|
|
|
| 39 |
"gpu": 2,
|
| 40 |
"status": "done",
|
| 41 |
"elapsed": 10932,
|
| 42 |
+
"idle_time": 4995,
|
| 43 |
"exit_code": 0,
|
| 44 |
"retries": 0,
|
| 45 |
"log_file": "/tmp/gpu_queue/job_002_as_sorl_abs30_K1_50K_ig1_1L3H5_gpu2.log"
|
|
|
|
| 51 |
"gpu": 0,
|
| 52 |
"status": "done",
|
| 53 |
"elapsed": 9082,
|
| 54 |
+
"idle_time": 6846,
|
| 55 |
"exit_code": 0,
|
| 56 |
"retries": 0,
|
| 57 |
"log_file": "/tmp/gpu_queue/job_003_as_sorl_abs30_K1_50K_ig5_2L1H1_gpu0.log"
|
|
|
|
| 63 |
"gpu": 1,
|
| 64 |
"status": "done",
|
| 65 |
"elapsed": 10997,
|
| 66 |
+
"idle_time": 4930,
|
| 67 |
"exit_code": 0,
|
| 68 |
"retries": 0,
|
| 69 |
"log_file": "/tmp/gpu_queue/job_004_as_sorl_abs30_K1_50K_ig5_1L3H5_gpu1.log"
|
|
|
|
| 75 |
"gpu": 2,
|
| 76 |
"status": "done",
|
| 77 |
"elapsed": 10836,
|
| 78 |
+
"idle_time": 5091,
|
| 79 |
"exit_code": 0,
|
| 80 |
"retries": 0,
|
| 81 |
"log_file": "/tmp/gpu_queue/job_005_as_sorl_abs30_K1_50K_abs05_2L1_gpu2.log"
|
|
|
|
| 122 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 30 --K 4 --nu",
|
| 123 |
"gpu": 0,
|
| 124 |
"status": "running",
|
| 125 |
+
"elapsed": 1309,
|
| 126 |
+
"idle_time": 7,
|
| 127 |
"exit_code": -1,
|
| 128 |
"retries": 0,
|
| 129 |
"log_file": "/tmp/gpu_queue/job_009_as_sorl_abs30_K4_100K_2L1H128d_gpu0.log"
|
|
|
|
| 146 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 100000 --abs_vocab 30 --K 4 --nu",
|
| 147 |
"gpu": 0,
|
| 148 |
"status": "running",
|
| 149 |
+
"elapsed": 3479,
|
| 150 |
+
"idle_time": 0,
|
| 151 |
"exit_code": -1,
|
| 152 |
"retries": 0,
|
| 153 |
"log_file": "/tmp/gpu_queue/job_011_as_sorl_abs30_K4_100K_1L3H510d_gpu0.log"
|
|
|
|
| 159 |
"gpu": 0,
|
| 160 |
"status": "done",
|
| 161 |
"elapsed": 1450,
|
| 162 |
+
"idle_time": 14457,
|
| 163 |
"exit_code": 0,
|
| 164 |
"retries": 0,
|
| 165 |
"log_file": "/tmp/gpu_queue/job_012_as_baseline_10K_2L3H510d_gpu0.log"
|
|
|
|
| 170 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 1 --K 1 --num_",
|
| 171 |
"gpu": 2,
|
| 172 |
"status": "running",
|
| 173 |
+
"elapsed": 4989,
|
| 174 |
+
"idle_time": 258,
|
| 175 |
"exit_code": -1,
|
| 176 |
"retries": 0,
|
| 177 |
"log_file": "/tmp/gpu_queue/job_013_as_sorl_abs1_K1_10K_2L3H510d_gpu2.log"
|
|
|
|
| 182 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 2 --K 1 --num_",
|
| 183 |
"gpu": 1,
|
| 184 |
"status": "running",
|
| 185 |
+
"elapsed": 4924,
|
| 186 |
+
"idle_time": 215,
|
| 187 |
"exit_code": -1,
|
| 188 |
"retries": 0,
|
| 189 |
"log_file": "/tmp/gpu_queue/job_014_as_sorl_abs2_K1_10K_2L3H510d_gpu1.log"
|
|
|
|
| 195 |
"gpu": 0,
|
| 196 |
"status": "done",
|
| 197 |
"elapsed": 5222,
|
| 198 |
+
"idle_time": 3485,
|
| 199 |
"exit_code": 0,
|
| 200 |
"retries": 0,
|
| 201 |
"log_file": "/tmp/gpu_queue/job_015_as_sorl_abs5_K1_10K_2L3H510d_gpu0.log"
|
|
|
|
| 206 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 10 --K 1 --num",
|
| 207 |
"gpu": 1,
|
| 208 |
"status": "running",
|
| 209 |
+
"elapsed": 5014,
|
| 210 |
+
"idle_time": 299,
|
| 211 |
"exit_code": -1,
|
| 212 |
"retries": 0,
|
| 213 |
"log_file": "/tmp/gpu_queue/job_016_as_sorl_abs10_K1_10K_2L3H510d_gpu1.log"
|
|
|
|
| 217 |
"name": "as_sorl_abs20_K1_10K_2L3H510d",
|
| 218 |
"cmd": "python -m arithmetic.train --mode sorl --ops add_sub --dataset_size 10000 --abs_vocab 20 --K 1 --num",
|
| 219 |
"gpu": 2,
|
| 220 |
+
"status": "done",
|
| 221 |
+
"elapsed": 5089,
|
| 222 |
+
"idle_time": 2,
|
| 223 |
+
"exit_code": 0,
|
| 224 |
"retries": 0,
|
| 225 |
"log_file": "/tmp/gpu_queue/job_017_as_sorl_abs20_K1_10K_2L3H510d_gpu2.log"
|
| 226 |
},
|
|
|
|
| 231 |
"gpu": 0,
|
| 232 |
"status": "done",
|
| 233 |
"elapsed": 5528,
|
| 234 |
+
"idle_time": 1314,
|
| 235 |
"exit_code": 0,
|
| 236 |
"retries": 0,
|
| 237 |
"log_file": "/tmp/gpu_queue/job_018_as_sorl_abs30_K1_10K_2L3H510d_gpu0.log"
|
|
|
|
| 243 |
"gpu": 0,
|
| 244 |
"status": "done",
|
| 245 |
"elapsed": 5747,
|
| 246 |
+
"idle_time": 8710,
|
| 247 |
"exit_code": 0,
|
| 248 |
"retries": 0,
|
| 249 |
"log_file": "/tmp/gpu_queue/job_019_as_sorl_abs50_K1_10K_2L3H510d_gpu0.log"
|