Upload queue_status.json with huggingface_hub
Browse files- queue_status.json +21 -21
queue_status.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
{
|
| 2 |
-
"timestamp": "2026-04-24 22:
|
| 3 |
"total": 4,
|
| 4 |
-
"pending":
|
| 5 |
"running": 2,
|
| 6 |
-
"done":
|
| 7 |
"failed": 0,
|
| 8 |
"stale": 0,
|
| 9 |
"retrying": 0,
|
|
@@ -13,10 +13,10 @@
|
|
| 13 |
"name": "PYTHONPATH=.",
|
| 14 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 1 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 15 |
"gpu": 0,
|
| 16 |
-
"status": "
|
| 17 |
-
"elapsed":
|
| 18 |
-
"idle_time":
|
| 19 |
-
"exit_code":
|
| 20 |
"retries": 1,
|
| 21 |
"log_file": "/tmp/stride_applications_queue/job_000_PYTHONPATH=._gpu0.log"
|
| 22 |
},
|
|
@@ -25,10 +25,10 @@
|
|
| 25 |
"name": "PYTHONPATH=.",
|
| 26 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 3 --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": 1,
|
| 33 |
"log_file": "/tmp/stride_applications_queue/job_001_PYTHONPATH=._gpu1.log"
|
| 34 |
},
|
|
@@ -36,25 +36,25 @@
|
|
| 36 |
"job_id": 2,
|
| 37 |
"name": "PYTHONPATH=.",
|
| 38 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 4 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 39 |
-
"gpu":
|
| 40 |
-
"status": "
|
| 41 |
-
"elapsed":
|
| 42 |
-
"idle_time":
|
| 43 |
"exit_code": -1,
|
| 44 |
"retries": 0,
|
| 45 |
-
"log_file": ""
|
| 46 |
},
|
| 47 |
{
|
| 48 |
"job_id": 3,
|
| 49 |
"name": "PYTHONPATH=.",
|
| 50 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 5 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 51 |
-
"gpu":
|
| 52 |
-
"status": "
|
| 53 |
-
"elapsed":
|
| 54 |
-
"idle_time":
|
| 55 |
"exit_code": -1,
|
| 56 |
"retries": 0,
|
| 57 |
-
"log_file": ""
|
| 58 |
}
|
| 59 |
]
|
| 60 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"timestamp": "2026-04-24 22:31:14",
|
| 3 |
"total": 4,
|
| 4 |
+
"pending": 0,
|
| 5 |
"running": 2,
|
| 6 |
+
"done": 2,
|
| 7 |
"failed": 0,
|
| 8 |
"stale": 0,
|
| 9 |
"retrying": 0,
|
|
|
|
| 13 |
"name": "PYTHONPATH=.",
|
| 14 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 1 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 15 |
"gpu": 0,
|
| 16 |
+
"status": "done",
|
| 17 |
+
"elapsed": 545,
|
| 18 |
+
"idle_time": 57,
|
| 19 |
+
"exit_code": 0,
|
| 20 |
"retries": 1,
|
| 21 |
"log_file": "/tmp/stride_applications_queue/job_000_PYTHONPATH=._gpu0.log"
|
| 22 |
},
|
|
|
|
| 25 |
"name": "PYTHONPATH=.",
|
| 26 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 3 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 27 |
"gpu": 1,
|
| 28 |
+
"status": "done",
|
| 29 |
+
"elapsed": 542,
|
| 30 |
+
"idle_time": 60,
|
| 31 |
+
"exit_code": 0,
|
| 32 |
"retries": 1,
|
| 33 |
"log_file": "/tmp/stride_applications_queue/job_001_PYTHONPATH=._gpu1.log"
|
| 34 |
},
|
|
|
|
| 36 |
"job_id": 2,
|
| 37 |
"name": "PYTHONPATH=.",
|
| 38 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 4 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 39 |
+
"gpu": 0,
|
| 40 |
+
"status": "running",
|
| 41 |
+
"elapsed": 55,
|
| 42 |
+
"idle_time": 37,
|
| 43 |
"exit_code": -1,
|
| 44 |
"retries": 0,
|
| 45 |
+
"log_file": "/tmp/stride_applications_queue/job_002_PYTHONPATH=._gpu0.log"
|
| 46 |
},
|
| 47 |
{
|
| 48 |
"job_id": 3,
|
| 49 |
"name": "PYTHONPATH=.",
|
| 50 |
"cmd": "PYTHONPATH=. python -m scripts.train_qwen --rate 0.01 --seed 5 --lr 5e-5 --proxy openwebtext/subset_20M_seed0.jsonl --re",
|
| 51 |
+
"gpu": 1,
|
| 52 |
+
"status": "running",
|
| 53 |
+
"elapsed": 55,
|
| 54 |
+
"idle_time": 37,
|
| 55 |
"exit_code": -1,
|
| 56 |
"retries": 0,
|
| 57 |
+
"log_file": "/tmp/stride_applications_queue/job_003_PYTHONPATH=._gpu1.log"
|
| 58 |
}
|
| 59 |
]
|
| 60 |
}
|