Upload queue_status.json with huggingface_hub
Browse files- queue_status.json +49 -49
queue_status.json
CHANGED
|
@@ -1,10 +1,10 @@
|
|
| 1 |
{
|
| 2 |
-
"timestamp": "2026-04-24 22:14:
|
| 3 |
"total": 12,
|
| 4 |
"pending": 0,
|
| 5 |
"running": 0,
|
| 6 |
-
"done":
|
| 7 |
-
"failed":
|
| 8 |
"stale": 0,
|
| 9 |
"retrying": 0,
|
| 10 |
"jobs": [
|
|
@@ -14,8 +14,8 @@
|
|
| 14 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 0 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 15 |
"gpu": 0,
|
| 16 |
"status": "done",
|
| 17 |
-
"elapsed":
|
| 18 |
-
"idle_time":
|
| 19 |
"exit_code": 0,
|
| 20 |
"retries": 0,
|
| 21 |
"log_file": "/tmp/stride_applications_queue/job_000_PYTHONPATH=._gpu0.log"
|
|
@@ -25,24 +25,24 @@
|
|
| 25 |
"name": "PYTHONPATH=.",
|
| 26 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 1 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 27 |
"gpu": 1,
|
| 28 |
-
"status": "
|
| 29 |
-
"elapsed":
|
| 30 |
-
"idle_time":
|
| 31 |
-
"exit_code":
|
| 32 |
-
"retries":
|
| 33 |
"log_file": "/tmp/stride_applications_queue/job_001_PYTHONPATH=._gpu1.log"
|
| 34 |
},
|
| 35 |
{
|
| 36 |
"job_id": 2,
|
| 37 |
"name": "PYTHONPATH=.",
|
| 38 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 2 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 39 |
-
"gpu":
|
| 40 |
"status": "failed",
|
| 41 |
-
"elapsed":
|
| 42 |
-
"idle_time":
|
| 43 |
-
"exit_code":
|
| 44 |
"retries": 1,
|
| 45 |
-
"log_file": "/tmp/stride_applications_queue/job_002_PYTHONPATH=.
|
| 46 |
},
|
| 47 |
{
|
| 48 |
"job_id": 3,
|
|
@@ -50,9 +50,9 @@
|
|
| 50 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 3 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 51 |
"gpu": 1,
|
| 52 |
"status": "failed",
|
| 53 |
-
"elapsed":
|
| 54 |
-
"idle_time":
|
| 55 |
-
"exit_code":
|
| 56 |
"retries": 1,
|
| 57 |
"log_file": "/tmp/stride_applications_queue/job_003_PYTHONPATH=._gpu1.log"
|
| 58 |
},
|
|
@@ -60,97 +60,97 @@
|
|
| 60 |
"job_id": 4,
|
| 61 |
"name": "PYTHONPATH=.",
|
| 62 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 4 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 63 |
-
"gpu":
|
| 64 |
"status": "failed",
|
| 65 |
-
"elapsed":
|
| 66 |
-
"idle_time":
|
| 67 |
"exit_code": 1,
|
| 68 |
"retries": 1,
|
| 69 |
-
"log_file": "/tmp/stride_applications_queue/job_004_PYTHONPATH=.
|
| 70 |
},
|
| 71 |
{
|
| 72 |
"job_id": 5,
|
| 73 |
"name": "PYTHONPATH=.",
|
| 74 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 5 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 75 |
-
"gpu":
|
| 76 |
"status": "failed",
|
| 77 |
"elapsed": 21,
|
| 78 |
-
"idle_time":
|
| 79 |
"exit_code": 1,
|
| 80 |
"retries": 1,
|
| 81 |
-
"log_file": "/tmp/stride_applications_queue/job_005_PYTHONPATH=.
|
| 82 |
},
|
| 83 |
{
|
| 84 |
"job_id": 6,
|
| 85 |
"name": "PYTHONPATH=.",
|
| 86 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 0 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 87 |
-
"gpu":
|
| 88 |
"status": "failed",
|
| 89 |
-
"elapsed":
|
| 90 |
-
"idle_time":
|
| 91 |
"exit_code": 1,
|
| 92 |
"retries": 1,
|
| 93 |
-
"log_file": "/tmp/stride_applications_queue/job_006_PYTHONPATH=.
|
| 94 |
},
|
| 95 |
{
|
| 96 |
"job_id": 7,
|
| 97 |
"name": "PYTHONPATH=.",
|
| 98 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 1 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 99 |
-
"gpu":
|
| 100 |
"status": "failed",
|
| 101 |
-
"elapsed":
|
| 102 |
-
"idle_time":
|
| 103 |
"exit_code": 1,
|
| 104 |
"retries": 1,
|
| 105 |
-
"log_file": "/tmp/stride_applications_queue/job_007_PYTHONPATH=.
|
| 106 |
},
|
| 107 |
{
|
| 108 |
"job_id": 8,
|
| 109 |
"name": "PYTHONPATH=.",
|
| 110 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 2 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 111 |
"gpu": 1,
|
| 112 |
-
"status": "
|
| 113 |
-
"elapsed":
|
| 114 |
-
"idle_time":
|
| 115 |
-
"exit_code":
|
| 116 |
-
"retries":
|
| 117 |
"log_file": "/tmp/stride_applications_queue/job_008_PYTHONPATH=._gpu1.log"
|
| 118 |
},
|
| 119 |
{
|
| 120 |
"job_id": 9,
|
| 121 |
"name": "PYTHONPATH=.",
|
| 122 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 3 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 123 |
-
"gpu":
|
| 124 |
"status": "failed",
|
| 125 |
-
"elapsed":
|
| 126 |
-
"idle_time":
|
| 127 |
"exit_code": 1,
|
| 128 |
"retries": 1,
|
| 129 |
-
"log_file": "/tmp/stride_applications_queue/job_009_PYTHONPATH=.
|
| 130 |
},
|
| 131 |
{
|
| 132 |
"job_id": 10,
|
| 133 |
"name": "PYTHONPATH=.",
|
| 134 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 4 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 135 |
-
"gpu":
|
| 136 |
"status": "failed",
|
| 137 |
"elapsed": 18,
|
| 138 |
-
"idle_time":
|
| 139 |
"exit_code": 1,
|
| 140 |
"retries": 1,
|
| 141 |
-
"log_file": "/tmp/stride_applications_queue/job_010_PYTHONPATH=.
|
| 142 |
},
|
| 143 |
{
|
| 144 |
"job_id": 11,
|
| 145 |
"name": "PYTHONPATH=.",
|
| 146 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 5 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 147 |
-
"gpu":
|
| 148 |
"status": "failed",
|
| 149 |
-
"elapsed":
|
| 150 |
-
"idle_time":
|
| 151 |
"exit_code": 1,
|
| 152 |
"retries": 1,
|
| 153 |
-
"log_file": "/tmp/stride_applications_queue/job_011_PYTHONPATH=.
|
| 154 |
}
|
| 155 |
]
|
| 156 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"timestamp": "2026-04-24 22:14:25",
|
| 3 |
"total": 12,
|
| 4 |
"pending": 0,
|
| 5 |
"running": 0,
|
| 6 |
+
"done": 1,
|
| 7 |
+
"failed": 11,
|
| 8 |
"stale": 0,
|
| 9 |
"retrying": 0,
|
| 10 |
"jobs": [
|
|
|
|
| 14 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 0 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 15 |
"gpu": 0,
|
| 16 |
"status": "done",
|
| 17 |
+
"elapsed": 555,
|
| 18 |
+
"idle_time": 1,
|
| 19 |
"exit_code": 0,
|
| 20 |
"retries": 0,
|
| 21 |
"log_file": "/tmp/stride_applications_queue/job_000_PYTHONPATH=._gpu0.log"
|
|
|
|
| 25 |
"name": "PYTHONPATH=.",
|
| 26 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 1 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 27 |
"gpu": 1,
|
| 28 |
+
"status": "failed",
|
| 29 |
+
"elapsed": 21,
|
| 30 |
+
"idle_time": 515,
|
| 31 |
+
"exit_code": 1,
|
| 32 |
+
"retries": 1,
|
| 33 |
"log_file": "/tmp/stride_applications_queue/job_001_PYTHONPATH=._gpu1.log"
|
| 34 |
},
|
| 35 |
{
|
| 36 |
"job_id": 2,
|
| 37 |
"name": "PYTHONPATH=.",
|
| 38 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 2 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 39 |
+
"gpu": 1,
|
| 40 |
"status": "failed",
|
| 41 |
+
"elapsed": 21,
|
| 42 |
+
"idle_time": 467,
|
| 43 |
+
"exit_code": 1,
|
| 44 |
"retries": 1,
|
| 45 |
+
"log_file": "/tmp/stride_applications_queue/job_002_PYTHONPATH=._gpu1.log"
|
| 46 |
},
|
| 47 |
{
|
| 48 |
"job_id": 3,
|
|
|
|
| 50 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 3 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 51 |
"gpu": 1,
|
| 52 |
"status": "failed",
|
| 53 |
+
"elapsed": 21,
|
| 54 |
+
"idle_time": 426,
|
| 55 |
+
"exit_code": 1,
|
| 56 |
"retries": 1,
|
| 57 |
"log_file": "/tmp/stride_applications_queue/job_003_PYTHONPATH=._gpu1.log"
|
| 58 |
},
|
|
|
|
| 60 |
"job_id": 4,
|
| 61 |
"name": "PYTHONPATH=.",
|
| 62 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 4 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 63 |
+
"gpu": 1,
|
| 64 |
"status": "failed",
|
| 65 |
+
"elapsed": 25,
|
| 66 |
+
"idle_time": 376,
|
| 67 |
"exit_code": 1,
|
| 68 |
"retries": 1,
|
| 69 |
+
"log_file": "/tmp/stride_applications_queue/job_004_PYTHONPATH=._gpu1.log"
|
| 70 |
},
|
| 71 |
{
|
| 72 |
"job_id": 5,
|
| 73 |
"name": "PYTHONPATH=.",
|
| 74 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 5 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 75 |
+
"gpu": 1,
|
| 76 |
"status": "failed",
|
| 77 |
"elapsed": 21,
|
| 78 |
+
"idle_time": 291,
|
| 79 |
"exit_code": 1,
|
| 80 |
"retries": 1,
|
| 81 |
+
"log_file": "/tmp/stride_applications_queue/job_005_PYTHONPATH=._gpu1.log"
|
| 82 |
},
|
| 83 |
{
|
| 84 |
"job_id": 6,
|
| 85 |
"name": "PYTHONPATH=.",
|
| 86 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 0 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 87 |
+
"gpu": 1,
|
| 88 |
"status": "failed",
|
| 89 |
+
"elapsed": 19,
|
| 90 |
+
"idle_time": 334,
|
| 91 |
"exit_code": 1,
|
| 92 |
"retries": 1,
|
| 93 |
+
"log_file": "/tmp/stride_applications_queue/job_006_PYTHONPATH=._gpu1.log"
|
| 94 |
},
|
| 95 |
{
|
| 96 |
"job_id": 7,
|
| 97 |
"name": "PYTHONPATH=.",
|
| 98 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 1 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 99 |
+
"gpu": 1,
|
| 100 |
"status": "failed",
|
| 101 |
+
"elapsed": 17,
|
| 102 |
+
"idle_time": 96,
|
| 103 |
"exit_code": 1,
|
| 104 |
"retries": 1,
|
| 105 |
+
"log_file": "/tmp/stride_applications_queue/job_007_PYTHONPATH=._gpu1.log"
|
| 106 |
},
|
| 107 |
{
|
| 108 |
"job_id": 8,
|
| 109 |
"name": "PYTHONPATH=.",
|
| 110 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 2 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 111 |
"gpu": 1,
|
| 112 |
+
"status": "failed",
|
| 113 |
+
"elapsed": 18,
|
| 114 |
+
"idle_time": 135,
|
| 115 |
+
"exit_code": 1,
|
| 116 |
+
"retries": 1,
|
| 117 |
"log_file": "/tmp/stride_applications_queue/job_008_PYTHONPATH=._gpu1.log"
|
| 118 |
},
|
| 119 |
{
|
| 120 |
"job_id": 9,
|
| 121 |
"name": "PYTHONPATH=.",
|
| 122 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 3 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 123 |
+
"gpu": 1,
|
| 124 |
"status": "failed",
|
| 125 |
+
"elapsed": 20,
|
| 126 |
+
"idle_time": 251,
|
| 127 |
"exit_code": 1,
|
| 128 |
"retries": 1,
|
| 129 |
+
"log_file": "/tmp/stride_applications_queue/job_009_PYTHONPATH=._gpu1.log"
|
| 130 |
},
|
| 131 |
{
|
| 132 |
"job_id": 10,
|
| 133 |
"name": "PYTHONPATH=.",
|
| 134 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 4 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 135 |
+
"gpu": 1,
|
| 136 |
"status": "failed",
|
| 137 |
"elapsed": 18,
|
| 138 |
+
"idle_time": 175,
|
| 139 |
"exit_code": 1,
|
| 140 |
"retries": 1,
|
| 141 |
+
"log_file": "/tmp/stride_applications_queue/job_010_PYTHONPATH=._gpu1.log"
|
| 142 |
},
|
| 143 |
{
|
| 144 |
"job_id": 11,
|
| 145 |
"name": "PYTHONPATH=.",
|
| 146 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 5 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 147 |
+
"gpu": 1,
|
| 148 |
"status": "failed",
|
| 149 |
+
"elapsed": 18,
|
| 150 |
+
"idle_time": 215,
|
| 151 |
"exit_code": 1,
|
| 152 |
"retries": 1,
|
| 153 |
+
"log_file": "/tmp/stride_applications_queue/job_011_PYTHONPATH=._gpu1.log"
|
| 154 |
}
|
| 155 |
]
|
| 156 |
}
|