scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-10 22:41:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 82

fix(anthropic-adapter): strip output_config for non-Anthropic backends

Browse source 
output_config is an Anthropic Claude-specific parameter that has no
equivalent in other model providers. When routing through the
experimental pass-through adapter to non-Anthropic backends (e.g.
Amazon Nova Pro on Bedrock, Llama, Mistral), the parameter is forwarded
verbatim and causes HTTP 400 "extraneous key" errors.

Strip output_config from the request when the target provider is not
an Anthropic Claude model. The thinking parameter is intentionally
excluded from stripping because some non-Anthropic models (e.g. Qwen)
support reasoning/thinking natively and the adapter already handles
translation for those cases.
This commit is contained in:
Rob Sherman 2026-03-03 17:33:36 -08:00
parent ea8ac80961
commit fd2f476d05
No known key found for this signature in database
1 changed files with 19 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

19
litellm/llms/anthropic/experimental_pass_through/adapters/handler.py
View file

@ -195,7 +195,26 @@ class LiteLLMMessagesToCompletionTransformationHandler:
"include_usage": True,
}
# These params are only understood by Anthropic Claude models.
# When routing to a non-Anthropic backend (e.g. Bedrock Nova Pro,
# Llama, Mistral), they are rejected as unknown fields. We strip
# them here so that the underlying model receives a clean request.
# Note: "thinking" is intentionally excluded from this list because
# some non-Anthropic models (e.g. Qwen) support reasoning/thinking
# and the existing adapter logic already handles translation for those.
_anthropic_only_params = {"output_config"}
_target_provider = (extra_kwargs or {}).get("custom_llm_provider", "")
_is_anthropic_claude = _target_provider in (
"anthropic",
) or (
_target_provider == "bedrock"
and "anthropic.claude" in completion_kwargs.get("model", "")
)
excluded_keys = {"anthropic_messages"}
if not _is_anthropic_claude:
excluded_keys = excluded_keys | _anthropic_only_params
extra_kwargs = extra_kwargs or {}
for key, value in extra_kwargs.items():
if (