Upload queue_status.json with huggingface_hub

queue_status.json

@@ -1,10 +1,10 @@
 {
-  "timestamp": "2026-04-25 17:35:36",
+  "timestamp": "2026-04-25 17:37:32",
   "total": 7,
   "pending": 0,
-  "running": 7,
+  "running": 0,
   "done": 0,
-  "failed": 0,
+  "failed": 7,
   "stale": 0,
   "retrying": 0,
   "jobs": [
@@ -13,10 +13,10 @@
       "name": "PYTHONPATH=.",
       "cmd": "PYTHONPATH=. /home/ubuntu/venvs/steering/bin/python -m scripts.train_qwen --mode contaminated --rate 0.01 --seed 0 --pro",
       "gpu": 0,
-      "status": "running",
-      "elapsed": 0,
-      "idle_time": 0,
-      "exit_code": -1,
+      "status": "failed",
+      "elapsed": 5,
+      "idle_time": 111,
+      "exit_code": 1,
       "retries": 1,
       "log_file": "/tmp/stride_applications_queue/job_000_PYTHONPATH=._gpu0.log"
     },
@@ -25,10 +25,10 @@
       "name": "PYTHONPATH=.",
       "cmd": "PYTHONPATH=. /home/ubuntu/venvs/steering/bin/python -m scripts.train_qwen --mode contaminated --rate 0.01 --seed 1 --pro",
       "gpu": 1,
-      "status": "running",
-      "elapsed": 0,
-      "idle_time": 0,
-      "exit_code": -1,
+      "status": "failed",
+      "elapsed": 5,
+      "idle_time": 112,
+      "exit_code": 1,
       "retries": 1,
       "log_file": "/tmp/stride_applications_queue/job_001_PYTHONPATH=._gpu1.log"
     },
@@ -37,11 +37,11 @@
       "name": "PYTHONPATH=.",
       "cmd": "PYTHONPATH=. /home/ubuntu/venvs/steering/bin/python -m scripts.train_qwen --mode contaminated --rate 0.01 --seed 2 --pro",
       "gpu": 2,
-      "status": "running",
+      "status": "failed",
       "elapsed": 6,
-      "idle_time": 6,
-      "exit_code": -1,
-      "retries": 0,
+      "idle_time": 109,
+      "exit_code": 1,
+      "retries": 1,
       "log_file": "/tmp/stride_applications_queue/job_002_PYTHONPATH=._gpu2.log"
     },
     {
@@ -49,11 +49,11 @@
       "name": "PYTHONPATH=.",
       "cmd": "PYTHONPATH=. /home/ubuntu/venvs/steering/bin/python -m scripts.train_qwen --mode contaminated --rate 0.01 --seed 3 --pro",
       "gpu": 3,
-      "status": "running",
-      "elapsed": 6,
-      "idle_time": 0,
-      "exit_code": -1,
-      "retries": 0,
+      "status": "failed",
+      "elapsed": 5,
+      "idle_time": 110,
+      "exit_code": 1,
+      "retries": 1,
       "log_file": "/tmp/stride_applications_queue/job_003_PYTHONPATH=._gpu3.log"
     },
     {
@@ -61,11 +61,11 @@
       "name": "PYTHONPATH=.",
       "cmd": "PYTHONPATH=. /home/ubuntu/venvs/steering/bin/python -m scripts.train_qwen --mode contaminated --rate 0.01 --seed 4 --pro",
       "gpu": 4,
-      "status": "running",
-      "elapsed": 6,
-      "idle_time": 6,
-      "exit_code": -1,
-      "retries": 0,
+      "status": "failed",
+      "elapsed": 5,
+      "idle_time": 109,
+      "exit_code": 1,
+      "retries": 1,
       "log_file": "/tmp/stride_applications_queue/job_004_PYTHONPATH=._gpu4.log"
     },
     {
@@ -73,24 +73,24 @@
       "name": "PYTHONPATH=.",
       "cmd": "PYTHONPATH=. /home/ubuntu/venvs/steering/bin/python -m scripts.train_qwen --mode contaminated --rate 0.01 --seed 5 --pro",
       "gpu": 5,
-      "status": "running",
+      "status": "failed",
       "elapsed": 6,
-      "idle_time": 0,
-      "exit_code": -1,
-      "retries": 0,
+      "idle_time": 110,
+      "exit_code": 1,
+      "retries": 1,
       "log_file": "/tmp/stride_applications_queue/job_005_PYTHONPATH=._gpu5.log"
     },
     {
       "job_id": 6,
       "name": "PYTHONPATH=.",
       "cmd": "PYTHONPATH=. /home/ubuntu/venvs/steering/bin/python -m scripts.train_qwen --mode clean --proxy openwebtext/subset_20M_se",
-      "gpu": 6,
-      "status": "running",
-      "elapsed": 6,
-      "idle_time": 6,
-      "exit_code": -1,
-      "retries": 0,
-      "log_file": "/tmp/stride_applications_queue/job_006_PYTHONPATH=._gpu6.log"
+      "gpu": 0,
+      "status": "failed",
+      "elapsed": 4,
+      "idle_time": 4,
+      "exit_code": -15,
+      "retries": 1,
+      "log_file": "/tmp/stride_applications_queue/job_006_PYTHONPATH=._gpu0.log"
     }
   ]
 }