Deploy audited RTX 6000D Flash-Next 128K multimodal baseline

scripts/monitor.py

@@ -0,0 +1,22 @@
+"""Log host/GPU headroom; stop the test server on host OOM or low RAM."""
+import json
+import subprocess
+import time
+from pathlib import Path
+
+def kv(path):
+    return {s.split()[0].rstrip(':'): int(s.split()[1]) for s in Path(path).read_text().splitlines() if len(s.split())>=2}
+
+baseline = kv('/proc/vmstat')['oom_kill']
+while True:
+    mem = kv('/proc/meminfo'); vm = kv('/proc/vmstat')
+    gpu = subprocess.run(['nvidia-smi','--query-gpu=memory.used,memory.total,utilization.gpu,temperature.gpu',
+                          '--format=csv,noheader,nounits'],capture_output=True,text=True,timeout=10)
+    low = mem['MemAvailable'] < 8*1024*1024
+    oom = vm['oom_kill'] > baseline
+    print(json.dumps(dict(time=time.time(),available_gib=mem['MemAvailable']/1024**2,
+                          oom_kill=vm['oom_kill'],gpu=gpu.stdout.strip(),stop=low or oom)),flush=True)
+    if low or oom:
+        subprocess.run(['docker','stop','-t','10','qwen38-flash-6000d'],timeout=30)
+        raise SystemExit('Host memory safeguard stopped server')
+    time.sleep(2)