scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-11-15 09:12:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 311

fix(bedrock): preserve restrictive request settings

Browse source
This commit is contained in:
Swayam Gupta 2026-09-04 19:26:30 +05:30
parent faa50f2660
commit ec6d2ae630
2 changed files with 40 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
litellm/llms/bedrock/chat/converse_transformation.py
View file

@ -467,7 +467,6 @@ class AmazonConverseConfig(BaseConfig):
max_tokens,
)
optional_params.pop("thinking", None)
optional_params.pop("output_config", None)
else:
optional_params["thinking"] = fitted_thinking
if fitted_thinking is not None and AnthropicConfig._is_adaptive_thinking_model(model, "bedrock"):
@ -894,10 +893,14 @@ class AmazonConverseConfig(BaseConfig):
drop_params: bool,
) -> dict:
is_thinking_enabled: Final = self.is_thinking_enabled(non_default_params)
max_tokens: Final = non_default_params.get("max_tokens")
max_completion_tokens: Final = non_default_params.get("max_completion_tokens")
resolved_max_tokens: Final = (
non_default_params.get("max_completion_tokens")
if non_default_params.get("max_completion_tokens") is not None
else non_default_params.get("max_tokens")
min(max_tokens, max_completion_tokens)
if isinstance(max_tokens, int) and isinstance(max_completion_tokens, int)
else max_completion_tokens
if max_completion_tokens is not None
else max_tokens
)
for param, value in non_default_params.items():

37
tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py
View file

@ -507,12 +507,12 @@ def test_reasoning_effort_budget_fits_max_tokens_for_anthropic_converse(
"params,expected_max_tokens,expects_thinking",
[
(
{"reasoning_effort": "medium", "max_tokens": 1024, "max_completion_tokens": 4096},
{"reasoning_effort": "medium", "max_tokens": 4096, "max_completion_tokens": 8192},
4096,
True,
),
(
{"reasoning_effort": "medium", "max_completion_tokens": 4096, "max_tokens": 1024},
{"reasoning_effort": "medium", "max_completion_tokens": 8192, "max_tokens": 4096},
4096,
True,
),
@ -528,10 +528,10 @@ def test_reasoning_effort_budget_fits_max_tokens_for_anthropic_converse(
),
],
)
def test_max_completion_tokens_wins_for_reasoning_effort_anthropic_converse(
def test_restrictive_token_alias_wins_for_reasoning_effort_anthropic_converse(
params, expected_max_tokens, expects_thinking
):
"""The completion-token alias controls both output and thinking limits regardless of input order."""
"""The smaller output-token alias controls both limits regardless of input order."""
config = AmazonConverseConfig()
model = "bedrock/converse/us.anthropic.claude-haiku-4-5-20251001-v1:0"
@ -549,6 +549,35 @@ def test_max_completion_tokens_wins_for_reasoning_effort_anthropic_converse(
assert "thinking" not in optional_params
@pytest.mark.parametrize("output_config_first", [True, False])
def test_unfitting_reasoning_effort_preserves_explicit_output_config(output_config_first):
"""Dropping generated thinking must not delete an independent structured-output configuration."""
config = AmazonConverseConfig()
model = "bedrock/converse/us.anthropic.claude-haiku-4-5-20251001-v1:0"
output_config = {"format": {"type": "json_schema", "schema": {"type": "object"}}}
params = {
"output_config": output_config,
"reasoning_effort": "medium",
"max_tokens": 1024,
}
if not output_config_first:
params = {
"reasoning_effort": "medium",
"max_tokens": 1024,
"output_config": output_config,
}
optional_params = config.map_openai_params(
non_default_params=params,
optional_params={},
model=model,
drop_params=False,
)
assert "thinking" not in optional_params
assert optional_params["output_config"] == output_config
def test_reasoning_effort_without_max_tokens_synthesizes_valid_limit():
"""Legacy thinking without an output limit receives enough completion-token headroom."""
from litellm.constants import DEFAULT_MAX_TOKENS