increase wait time for llm backends

lib/backend.py

@@ -30,7 +30,7 @@ from lib.data_types import (
     BenchmarkResult
 )
 
-VERSION = "0.2.0"
+VERSION = "0.1.0"
 
 MSG_HISTORY_LEN = 100
 log = logging.getLogger(__file__)
@@ -69,14 +69,10 @@ class Backend:
     report_addr: str = dataclasses.field(
         default_factory=lambda: os.environ.get("REPORT_ADDR", "https://run.vast.ai")
     )
-    mtoken: str = dataclasses.field(
-        default_factory=lambda: os.environ.get("MASTER_TOKEN", "")
-    )
 
     def __post_init__(self):
         self.metrics = Metrics()
         self.metrics._set_version(self.version)
-        self.metrics._set_mtoken(self.mtoken)
         self._total_pubkey_fetch_errors = 0
         self._pubkey = self._fetch_pubkey()
         self.__start_healthcheck: bool = False

lib/data_types.py

@@ -286,7 +286,6 @@ class AutoScalerData:
     """Data that is reported to autoscaler"""
 
     id: int
-    mtoken: str
     version: str
     loadtime: float
     cur_load: float

lib/metrics.py

@@ -28,7 +28,6 @@ def get_url() -> str:
 @dataclass
 class Metrics:
     version: str = "0"
-    mtoken: str = ""
     last_metric_update: float = 0.0
     last_request_served: float = 0.0
     update_pending: bool = False
@@ -143,16 +142,12 @@ class Metrics:
     def _set_version(self, version: str) -> None:
         self.version = version
 
-    def _set_mtoken(self, mtoken: str) -> None:
-        self.mtoken = mtoken
-
     #######################################Private#######################################
 
     async def __send_delete_requests_and_reset(self):
         async def post(report_addr: str, idxs: list[int], success_flag: bool) -> bool:
             data = {
                 "worker_id": self.id,
-                "mtoken": self.mtoken,
                 "request_idxs": idxs,
                 "success": success_flag,
             }
@@ -214,7 +209,6 @@ class Metrics:
         def compute_autoscaler_data() -> AutoScalerData:
             return AutoScalerData(
                 id=self.id,
-                mtoken=self.mtoken,
                 version=self.version,
                 loadtime=(loadtime_snapshot or 0.0), 
                 new_load=self.model_metrics.workload_processing,
@@ -234,25 +228,17 @@ class Metrics:
 
         async def send_data(report_addr: str) -> bool:
             data = compute_autoscaler_data()
-            log_data = asdict(data)
-            def obfuscate(secret: str) -> str:
-                if secret is None:
-                    return ""
-                return secret[:7] + "..." if len(secret) > 7 else ("*" * len(secret))
-            
-            log_data["mtoken"] = obfuscate(log_data.get("mtoken"))
+            full_path = report_addr.rstrip("/") + "/worker_status/"
             log.debug(
                 "\n".join(
                     [
                         "#" * 60,
                         f"sending data to autoscaler",
-                        f"{json.dumps(log_data, indent=2)}",
+                        f"{json.dumps((asdict(data)), indent=2)}",
                         "#" * 60,
                     ]
                 )
             )
-
-            full_path = report_addr.rstrip("/") + "/worker_status/"
             for attempt in range(1, 4):
                 try:
                     session = await self.http()

workers/openai/server.py

@@ -43,6 +43,7 @@ backend = Backend(
             for error_msg in MODEL_SERVER_ERROR_LOG_MSGS
         ],
     ],
+    max_wait_time=600
 )
 
 

workers/tgi/server.py

@@ -113,6 +113,7 @@ backend = Backend(
             for error_msg in MODEL_SERVER_ERROR_LOG_MSGS
         ],
     ],
+    max_wait_time=600
 )