server: fixes initial timeout for llm warmup

server/reflector/llm/base.py

@@ -35,7 +35,7 @@ class LLM:
         name = self.__class__.__name__
         logger.info(f"LLM[{name}] warming up...")
         try:
-            await retry(self._warmup)(logger=logger)
+            await self._warmup(logger=logger)
             duration = monotonic() - start
             logger.info(f"LLM[{name}] warmup took {duration:.2f} seconds")
         except Exception:

server/reflector/llm/llm_modal.py

@@ -19,7 +19,7 @@ class ModalLLM(LLM):
             response = await client.post(
                 self.llm_warmup_url,
                 headers=self.headers,
-                timeout=self.timeout,
+                timeout=60**5,
             )
             response.raise_for_status()