scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-14 23:21:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 156

fix: increase health check max_tokens from 5 to 16 (#23836)

Browse source 
GPT-5 models enforce a minimum of 16 for max_output_tokens. The current
default of 5 still causes health checks to fail for these models. Bump
the non-wildcard default to 16 — the smallest value that satisfies all
known provider minimums while keeping health checks lightweight.

Also tightens the wildcard test assertion from a weak disjunctive check
to strict key-absence.
This commit is contained in:
Hannah Smith 2026-04-27 11:56:47 -04:00
parent 3d2b8fed32
commit fd3eadbd2e
2 changed files with 8 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
litellm/proxy/health_check.py
View file

@ -319,7 +319,7 @@ def _resolve_health_check_max_tokens(
3. For non-wildcard reasoning routes: BACKGROUND_HEALTH_CHECK_MAX_TOKENS_REASONING
from env (if set)
4. BACKGROUND_HEALTH_CHECK_MAX_TOKENS (global, any route including wildcards)
5. Non-wildcard default: 5
5. Non-wildcard default: 16
6. Wildcard and nothing from (1)(4): leave unset (caller omits max_tokens)
"""
explicit = model_info.get("health_check_max_tokens", None)
@ -350,7 +350,7 @@ def _resolve_health_check_max_tokens(
return int(BACKGROUND_HEALTH_CHECK_MAX_TOKENS)
if not is_wildcard:
return 5
return 16
return None

13
tests/test_litellm/proxy/test_health_check_max_tokens.py
View file

@ -13,7 +13,7 @@ from litellm.proxy.health_check import (
@pytest.mark.asyncio
async def test_update_litellm_params_max_tokens_default(monkeypatch):
"""
Test that max_tokens defaults to 5 for non-wildcard models.
Test that max_tokens defaults to 16 for non-wildcard models.
"""
monkeypatch.setattr(hc_module, "BACKGROUND_HEALTH_CHECK_MAX_TOKENS", None)
monkeypatch.setattr(hc_module, "BACKGROUND_HEALTH_CHECK_MAX_TOKENS_REASONING", None)
@ -22,7 +22,7 @@ async def test_update_litellm_params_max_tokens_default(monkeypatch):
updated_params = _update_litellm_params_for_health_check(model_info, litellm_params)
assert updated_params["max_tokens"] == 5
assert updated_params["max_tokens"] == 16
@pytest.mark.asyncio
@ -48,8 +48,7 @@ async def test_update_litellm_params_max_tokens_wildcard():
updated_params = _update_litellm_params_for_health_check(model_info, litellm_params)
# Should not be set to 1
assert "max_tokens" not in updated_params or updated_params["max_tokens"] != 1
assert "max_tokens" not in updated_params
@pytest.mark.asyncio
@ -160,14 +159,14 @@ def test_explicit_health_check_max_tokens_beats_reasoning_specific():
def test_reasoning_specific_falls_through_when_wrong_branch_only(monkeypatch):
"""Only non-reasoning key set but model is reasoning → fall back to default 5."""
"""Only non-reasoning key set but model is reasoning → fall back to default 16."""
monkeypatch.setattr(hc_module, "BACKGROUND_HEALTH_CHECK_MAX_TOKENS", None)
monkeypatch.setattr(hc_module, "BACKGROUND_HEALTH_CHECK_MAX_TOKENS_REASONING", None)
model_info = {"health_check_max_tokens_non_reasoning": 3}
litellm_params = {"model": "openai/o1"}
with patch.object(hc_module.litellm, "supports_reasoning", return_value=True):
assert _resolve_health_check_max_tokens(model_info, litellm_params) == 5
assert _resolve_health_check_max_tokens(model_info, litellm_params) == 16
@pytest.mark.asyncio
@ -180,7 +179,7 @@ async def test_background_split_env_reasoning_vs_non_reasoning(monkeypatch):
with patch.object(hc_module.litellm, "supports_reasoning", return_value=False):
updated = _update_litellm_params_for_health_check(model_info, litellm_params)
assert updated["max_tokens"] == 5
assert updated["max_tokens"] == 16
litellm_params2 = {"model": "openai/o1"}
with patch.object(hc_module.litellm, "supports_reasoning", return_value=True):