Report usage for beam search (#6404)

vllm/sampling_params.py

@@ -189,6 +189,10 @@ class SamplingParams:
 
         self._verify_args()
         if self.use_beam_search:
+            # Lazy import to avoid circular imports.
+            from vllm.usage.usage_lib import set_runtime_usage_data
+            set_runtime_usage_data("use_beam_search", True)
+
             if not envs.VLLM_NO_DEPRECATION_WARNING:
                 logger.warning(
                     "[IMPORTANT] We plan to discontinue the support for beam "
@@ -196,6 +200,7 @@ class SamplingParams:
                     "https://github.com/vllm-project/vllm/issues/6226 for "
                     "more information. Set VLLM_NO_DEPRECATION_WARNING=1 to "
                     "suppress this warning.")
+
             self._verify_beam_search()
         else:
             self._verify_non_beam_search()

vllm/usage/usage_lib.py

@@ -7,7 +7,7 @@ import time
 from enum import Enum
 from pathlib import Path
 from threading import Thread
-from typing import Any, Dict, Optional
+from typing import Any, Dict, Optional, Union
 from uuid import uuid4
 
 import cpuinfo
@@ -25,6 +25,13 @@ _USAGE_STATS_DO_NOT_TRACK_PATH = os.path.join(_config_home,
 _USAGE_STATS_ENABLED = None
 _USAGE_STATS_SERVER = envs.VLLM_USAGE_STATS_SERVER
 
+_GLOBAL_RUNTIME_DATA: Dict[str, Union[str, int, bool]] = {}
+
+
+def set_runtime_usage_data(key: str, value: Union[str, int, bool]) -> None:
+    """Set global usage data that will be sent with every usage heartbeat."""
+    _GLOBAL_RUNTIME_DATA[key] = value
+
 
 def is_usage_stats_enabled():
     """Determine whether or not we can send usage stats to the server.
@@ -187,7 +194,11 @@ class UsageMessage:
         """
         while True:
             time.sleep(600)
-            data = {"uuid": self.uuid, "log_time": _get_current_timestamp_ns()}
+            data = {
+                "uuid": self.uuid,
+                "log_time": _get_current_timestamp_ns(),
+            }
+            data.update(_GLOBAL_RUNTIME_DATA)
 
             self._write_to_file(data)
             self._send_to_server(data)