Upload queue_status.json with huggingface_hub
Browse files- queue_status.json +7 -7
queue_status.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"timestamp": "2026-04-24 22:
|
| 3 |
"total": 4,
|
| 4 |
"pending": 2,
|
| 5 |
"running": 2,
|
|
@@ -14,10 +14,10 @@
|
|
| 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": "running",
|
| 17 |
-
"elapsed":
|
| 18 |
-
"idle_time":
|
| 19 |
"exit_code": -1,
|
| 20 |
-
"retries":
|
| 21 |
"log_file": "/tmp/stride_applications_queue/job_000_PYTHONPATH=._gpu0.log"
|
| 22 |
},
|
| 23 |
{
|
|
@@ -26,10 +26,10 @@
|
|
| 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": "running",
|
| 29 |
-
"elapsed":
|
| 30 |
-
"idle_time":
|
| 31 |
"exit_code": -1,
|
| 32 |
-
"retries":
|
| 33 |
"log_file": "/tmp/stride_applications_queue/job_001_PYTHONPATH=._gpu1.log"
|
| 34 |
},
|
| 35 |
{
|
|
|
|
| 1 |
{
|
| 2 |
+
"timestamp": "2026-04-24 22:23:12",
|
| 3 |
"total": 4,
|
| 4 |
"pending": 2,
|
| 5 |
"running": 2,
|
|
|
|
| 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": "running",
|
| 17 |
+
"elapsed": 119,
|
| 18 |
+
"idle_time": 103,
|
| 19 |
"exit_code": -1,
|
| 20 |
+
"retries": 1,
|
| 21 |
"log_file": "/tmp/stride_applications_queue/job_000_PYTHONPATH=._gpu0.log"
|
| 22 |
},
|
| 23 |
{
|
|
|
|
| 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": "running",
|
| 29 |
+
"elapsed": 119,
|
| 30 |
+
"idle_time": 103,
|
| 31 |
"exit_code": -1,
|
| 32 |
+
"retries": 1,
|
| 33 |
"log_file": "/tmp/stride_applications_queue/job_001_PYTHONPATH=._gpu1.log"
|
| 34 |
},
|
| 35 |
{
|