chore(lilith_service_fastapi): 🔧 Update FastAPI lifespan configuration in lifespan.py

Co-Authored-By: Lilith Autocommit <noreply@atlilith.com>

src/lilith_service_fastapi_bootstrap/lifespan.py

@@ -384,7 +384,10 @@ class GPULifespanManager(LifespanManager):
             self._boss = None
             return
 
-        # Watchdog is now automatic in v3.0+ - no manual start needed
+        # Auto-detect and register GPUs if not already registered
+        gpu_count = await self._boss.get_gpu_count()
+        if gpu_count == 0:
+            await self._register_gpus()
 
         # Create managed loaders (unified in v3.0+)
         self._gguf_loader = ManagedModelLoader(boss=self._boss)
@@ -412,3 +415,47 @@ class GPULifespanManager(LifespanManager):
             await self._boss.close()
 
         logger.info("GPU resources cleaned up")
+
+    async def _register_gpus(self) -> None:
+        """Auto-detect NVIDIA GPUs and register them with GPUBoss.
+
+        Uses nvidia-smi to query GPU names and VRAM. Falls back gracefully
+        if nvidia-smi is not available (e.g., CPU-only environments).
+        """
+        import shutil
+        import subprocess
+
+        if not shutil.which("nvidia-smi"):
+            logger.warning(
+                "nvidia-smi not found — cannot auto-detect GPUs. "
+                "GPU lease coordination requires manual registration."
+            )
+            return
+
+        try:
+            result = subprocess.run(
+                [
+                    "nvidia-smi",
+                    "--query-gpu=index,name,memory.total",
+                    "--format=csv,noheader,nounits",
+                ],
+                capture_output=True,
+                text=True,
+                timeout=10,
+            )
+
+            if result.returncode != 0:
+                logger.warning(f"nvidia-smi failed: {result.stderr.strip()}")
+                return
+
+            for line in result.stdout.strip().split("\n"):
+                parts = [p.strip() for p in line.split(",")]
+                if len(parts) != 3:
+                    continue
+                gpu_index = int(parts[0])
+                gpu_name = parts[1]
+                vram_total_mb = int(float(parts[2]))
+                await self._boss.initialize_gpu(gpu_index, vram_total_mb, gpu_name)
+
+        except (subprocess.TimeoutExpired, ValueError, OSError) as e:
+            logger.warning(f"GPU auto-detection failed: {e}")