diff --git a/src/art/client.py b/src/art/client.py
index fe7dbdde..a45d1278 100644
--- a/src/art/client.py
+++ b/src/art/client.py
@@ -1,12 +1,18 @@
 import os
-from typing import Any, Iterable, Literal, TypedDict, cast
+from typing import Any, Iterable, Literal, Type, TypedDict, TypeVar, cast
 
 import httpx
 from openai import AsyncOpenAI, BaseModel, _exceptions
-from openai._base_client import AsyncAPIClient, AsyncPaginator, make_request_options
+from openai._base_client import (
+    AsyncAPIClient,
+    AsyncPaginator,
+    FinalRequestOptions,
+    make_request_options,
+)
 from openai._compat import cached_property
 from openai._qs import Querystring
 from openai._resource import AsyncAPIResource
+from openai._streaming import AsyncStream
 from openai._types import NOT_GIVEN, NotGiven, Omit
 from openai._utils import is_mapping, maybe_transform
 from openai._version import __version__
@@ -17,6 +23,8 @@
 
 from .trajectories import TrajectoryGroup
 
+ResponseT = TypeVar("ResponseT")
+
 
 class Model(BaseModel):
     id: str
@@ -214,7 +222,23 @@ def __init__(
             version=__version__,
             base_url=base_url or "https://api.training.wandb.ai/v1",
             _strict_response_validation=False,
-            max_retries=0,
+            max_retries=3,
+        )
+
+    @override
+    async def request(
+        self,
+        cast_to: Type[ResponseT],
+        options: FinalRequestOptions,
+        *,
+        stream: bool = False,
+        stream_cls: type[AsyncStream[Any]] | None = None,
+    ) -> ResponseT | AsyncStream[Any]:
+        # Disable retries for POST requests
+        if options.method.upper() == "POST":
+            options.max_retries = 0
+        return await super().request(
+            cast_to=cast_to, options=options, stream=stream, stream_cls=stream_cls
         )
 
     @cached_property
diff --git a/src/art/serverless/backend.py b/src/art/serverless/backend.py
index 8ca957bf..604faea5 100644
--- a/src/art/serverless/backend.py
+++ b/src/art/serverless/backend.py
@@ -136,7 +136,7 @@ async def _train_model(
         num_sequences: int | None = None
         pbar: tqdm.tqdm | None = None
         while True:
-            await asyncio.sleep(0.5)
+            await asyncio.sleep(1)
             async for event in self._client.training_jobs.events.list(
                 training_job_id=training_job.id, after=after or NOT_GIVEN
             ):