Report accelerator type in /v1/loads (#32348)

Co-authored-by: Yinghai Lu <yinghai@meta.com>

python/sglang/srt/entrypoints/v1_loads.py

@@ -20,16 +20,24 @@ metrics for load balancing, monitoring, and capacity planning.
 
 import time
 from datetime import datetime, timezone
+from functools import lru_cache
 from typing import Optional
 
 from fastapi import APIRouter, Depends, HTTPException
 from fastapi.responses import Response
 
+from sglang.srt.utils import get_device_name
 from sglang.version import __version__
 
 router = APIRouter()
 
 
+@lru_cache(maxsize=1)
+def _accelerator_name() -> Optional[str]:
+    """Accelerator marketing name (e.g. "NVIDIA GB300"), None if unavailable."""
+    return get_device_name()
+
+
 def _get_tokenizer_manager():
     """Dependency to get tokenizer_manager from global state."""
     from sglang.srt.entrypoints.http_server import get_global_state
@@ -87,7 +95,7 @@ async def get_loads(
         format: Response format - 'json' (default) or 'prometheus'
 
     Returns:
-        JSON response with timestamp, version, and per-DP-rank loads
+        JSON response with timestamp, version, accelerator, and per-DP-rank loads
     """
     include_list = [s.strip() for s in include.split(",")] if include else None
 
@@ -119,5 +127,6 @@ async def get_loads(
     return {
         "timestamp": datetime.now(timezone.utc).isoformat(),
         "version": __version__,
+        "accelerator": _accelerator_name(),
         "loads": loads,
     }