scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-11-13 08:52:02 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 174

fix(exceptions): classify SGLang context window overflow

Browse source 
Co-authored-by: openhands <openhands@all-hands.dev>
This commit is contained in:
openhands 2026-10-28 14:26:51 +00:00
parent 38b4d31447
commit 4fcb5ecf5a
2 changed files with 62 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
litellm/litellm_core_utils/exception_mapping_utils.py
View file

@ -95,6 +95,7 @@ class ExceptionCheckers:
known_exception_substrings: Final = [
"exceed context limit",
"this model's maximum context length is",
"requested token count exceeds the model's maximum context length",
"string too long. expected a string with maximum length",
"model's maximum context limit",
"is longer than the model's context length",

62
tests/unit/litellm_core_utils/test_exception_mapping_utils.py
View file

@ -58,7 +58,6 @@ def test_openai_cyber_policy_uses_the_structured_error_code(shape: str, code: st
assert caught.value.response.headers["x-request-id"] == "cyber-policy-request"
def _debug_info_for_api_base(configured_api_base: str) -> str:
request: Final = httpx.Request("POST", "https://gateway.example/v1/chat/completions")
original: Final = OpenAIError(
@ -439,6 +438,67 @@ def test_gemini_context_window_error_mapping(error_message, should_raise_context
)
@pytest.mark.parametrize("sync_mode", [True, False])
@pytest.mark.parametrize("streaming", [True, False])
@pytest.mark.parametrize(
("error_message", "expected_type"),
[
(
(
"Requested token count exceeds the model's maximum context length of 131072 tokens. "
"You requested a total of 140000 tokens: 130000 tokens from the input messages and "
"10000 tokens for the completion. Please reduce the number of tokens in the input "
"messages or the completion to fit within the limit."
),
litellm.ContextWindowExceededError,
),
(
(
"requested token count exceeds the model's maximum context length of 8192 tokens. "
"you requested a total of 9000 tokens"
),
litellm.ContextWindowExceededError,
),
("max_tokens must be a positive integer", litellm.BadRequestError),
],
ids=["sglang-context-overflow", "lowercase-context-overflow", "unrelated-bad-request"],
)
@pytest.mark.asyncio
async def test_sglang_context_window_error_mapping(
sync_mode: bool,
streaming: bool,
error_message: str,
expected_type: type[litellm.BadRequestError],
) -> None:
def reject_request(request: httpx.Request) -> httpx.Response:
return httpx.Response(
400,
request=request,
json={"error": {"message": error_message, "type": "BadRequestError", "param": None, "code": 400}},
)
transport: Final = httpx.MockTransport(reject_request)
if sync_mode:
with (
httpx.Client(transport=transport) as http_client,
OpenAI(api_key="sk-test", max_retries=0, http_client=http_client) as client,
pytest.raises(litellm.BadRequestError) as caught,
):
_call_openai_api_sync("openai/sglang-model", "chat_completion", streaming, client)
else:
async with (
httpx.AsyncClient(transport=transport) as async_http_client,
AsyncOpenAI(api_key="sk-test", max_retries=0, http_client=async_http_client) as async_client,
):
with pytest.raises(litellm.BadRequestError) as caught:
await _call_openai_api_async("openai/sglang-model", "chat_completion", streaming, async_client)
assert type(caught.value) is expected_type
assert caught.value.status_code == 400
assert error_message in str(caught.value)
assert caught.value.llm_provider == "openai"
def test_lemonade_context_window_error_mapping():
"""Lemonade's llama.cpp backend should map context overflows to LiteLLM's standard error."""