From 326e2bb0bf460ddd93c142f2d08fff7f2636f05c Mon Sep 17 00:00:00 2001 From: jhpiedrahitao Date: Mon, 5 Oct 2026 15:37:39 -0500 Subject: [PATCH] feat(sambanova): support more chat params and restore X-Integration-Source header --- litellm/llms/sambanova/chat.py | 69 +++++++++-- .../test_sambanova_chat_transformation.py | 111 ++++++++++++++++++ 2 files changed, 171 insertions(+), 9 deletions(-) create mode 100644 tests/unit/llms/sambanova/test_sambanova_chat_transformation.py diff --git a/litellm/llms/sambanova/chat.py b/litellm/llms/sambanova/chat.py index 0e7dadf7062..7fa281c7120 100644 --- a/litellm/llms/sambanova/chat.py +++ b/litellm/llms/sambanova/chat.py @@ -11,8 +11,12 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( handle_messages_with_content_list_to_str_conversion, ) from litellm.llms.openai.chat.gpt_transformation import OpenAIGPTConfig +from litellm.secret_managers.main import get_secret_str from litellm.types.llms.openai import AllMessageValues +DEFAULT_INTEGRATION_SOURCE: Final = "litellm" +INTEGRATION_SOURCE_HEADER: Final = "X-Integration-Source" + class SambanovaConfig(OpenAIGPTConfig): """ @@ -59,9 +63,9 @@ class SambanovaConfig(OpenAIGPTConfig): Get the supported OpenAI params for the given model """ - from litellm.utils import supports_function_calling + from litellm.utils import supports_function_calling, supports_reasoning - params: Final = [ + params: Final = ( "max_completion_tokens", "max_tokens", "response_format", @@ -71,14 +75,24 @@ class SambanovaConfig(OpenAIGPTConfig): "temperature", "top_p", "top_k", - ] + "presence_penalty", + "frequency_penalty", + "logprobs", + "top_logprobs", + "n", + "logit_bias", + "seed", + ) + tool_params: Final = ( + ("tools", "tool_choice", "parallel_tool_calls") + if supports_function_calling(model, custom_llm_provider="sambanova") + else () + ) + reasoning_params: Final = ( + ("reasoning_effort",) if supports_reasoning(model, custom_llm_provider="sambanova") else () + ) - if supports_function_calling(model, custom_llm_provider="sambanova"): - params.append("tools") - params.append("tool_choice") - params.append("parallel_tool_calls") - - return params + return [*params, *tool_params, *reasoning_params] def map_openai_params( self, @@ -98,6 +112,43 @@ class SambanovaConfig(OpenAIGPTConfig): optional_params[param] = value return optional_params + def transform_request( + self, + model: str, + messages: list[AllMessageValues], + optional_params: dict, + litellm_params: dict, + headers: dict, + ) -> dict: + """ + Add the X-Integration-Source header SambaNova uses for analytics. + + Priority: `extra_headers["X-Integration-Source"]` > `extra_body["integration_source"]` + > SAMBANOVA_INTEGRATION_SOURCE env var > "litellm". + """ + extra_headers: Final[dict[str, str]] = dict(optional_params.get("extra_headers") or {}) + extra_body: Final[dict[str, object]] = dict(optional_params.get("extra_body") or {}) + integration_source: Final[str] = ( + extra_headers.get(INTEGRATION_SOURCE_HEADER) + or extra_body.pop("integration_source", None) + or get_secret_str("SAMBANOVA_INTEGRATION_SOURCE") + or DEFAULT_INTEGRATION_SOURCE + ) + other_params: Final[dict[str, object]] = {k: v for k, v in optional_params.items() if k != "extra_body"} + updated_params: Final[dict[str, object]] = { + **other_params, + "extra_headers": {**extra_headers, INTEGRATION_SOURCE_HEADER: integration_source}, + **({"extra_body": extra_body} if extra_body else {}), + } + + return super().transform_request( + model=model, + messages=messages, + optional_params=updated_params, + litellm_params=litellm_params, + headers=headers, + ) + @overload def _transform_messages( self, messages: list[AllMessageValues], model: str, is_async: Literal[True] diff --git a/tests/unit/llms/sambanova/test_sambanova_chat_transformation.py b/tests/unit/llms/sambanova/test_sambanova_chat_transformation.py new file mode 100644 index 00000000000..bd8d39bcfb5 --- /dev/null +++ b/tests/unit/llms/sambanova/test_sambanova_chat_transformation.py @@ -0,0 +1,111 @@ +import pytest + +from litellm.llms.sambanova.chat import SambanovaConfig +from litellm.utils import get_optional_params + +# Params documented in SambaNova's chat completions API spec that map 1:1 from OpenAI. +_ALWAYS_SUPPORTED = ( + "max_tokens", + "max_completion_tokens", + "temperature", + "top_p", + "top_k", + "stop", + "stream", + "stream_options", + "response_format", + "presence_penalty", + "frequency_penalty", + "logprobs", + "top_logprobs", + "n", + "logit_bias", + "seed", +) + + +@pytest.mark.parametrize("param", _ALWAYS_SUPPORTED) +def test_supported_params_include_spec_params(param): + assert param in SambanovaConfig().get_supported_openai_params("MiniMax-M3") + + +def test_reasoning_effort_only_for_reasoning_models(): + config = SambanovaConfig() + assert "reasoning_effort" in config.get_supported_openai_params("gpt-oss-120b") + assert "reasoning_effort" not in config.get_supported_openai_params("Meta-Llama-3.3-70B-Instruct") + + +def test_tools_only_for_function_calling_models(): + config = SambanovaConfig() + for param in ("tools", "tool_choice", "parallel_tool_calls"): + assert param in config.get_supported_openai_params("MiniMax-M3") + assert param not in config.get_supported_openai_params("gemma-4-31B-it") + + +def test_max_completion_tokens_mapped_to_max_tokens(): + optional_params = SambanovaConfig().map_openai_params( + non_default_params={"max_completion_tokens": 128}, + optional_params={}, + model="MiniMax-M3", + drop_params=False, + ) + assert optional_params == {"max_tokens": 128} + + +def test_new_params_forwarded_via_get_optional_params(): + optional_params = get_optional_params( + model="MiniMax-M3", + custom_llm_provider="sambanova", + seed=7, + presence_penalty=0.5, + frequency_penalty=0.25, + reasoning_effort="low", + ) + assert optional_params["seed"] == 7 + assert optional_params["presence_penalty"] == 0.5 + assert optional_params["frequency_penalty"] == 0.25 + assert optional_params["reasoning_effort"] == "low" + + +def _transform(optional_params): + return SambanovaConfig().transform_request( + model="MiniMax-M3", + messages=[{"role": "user", "content": "hi"}], + optional_params=optional_params, + litellm_params={}, + headers={}, + ) + + +def test_integration_source_defaults_to_litellm(monkeypatch): + monkeypatch.delenv("SAMBANOVA_INTEGRATION_SOURCE", raising=False) + request = _transform({}) + assert request["extra_headers"]["X-Integration-Source"] == "litellm" + + +def test_integration_source_env_overrides_default(monkeypatch): + monkeypatch.setenv("SAMBANOVA_INTEGRATION_SOURCE", "langflow") + request = _transform({}) + assert request["extra_headers"]["X-Integration-Source"] == "langflow" + + +def test_integration_source_param_overrides_env(monkeypatch): + monkeypatch.setenv("SAMBANOVA_INTEGRATION_SOURCE", "langflow") + optional_params = {"extra_body": {"integration_source": "my-framework"}} + request = _transform(optional_params) + assert request["extra_headers"]["X-Integration-Source"] == "my-framework" + assert "integration_source" not in request.get("extra_body", {}) + assert "integration_source" in optional_params["extra_body"], "caller's dict must not be mutated" + + +def test_integration_source_explicit_header_wins(monkeypatch): + monkeypatch.setenv("SAMBANOVA_INTEGRATION_SOURCE", "langflow") + request = _transform({"extra_headers": {"X-Integration-Source": "custom", "X-Other": "1"}}) + assert request["extra_headers"] == {"X-Integration-Source": "custom", "X-Other": "1"} + + +def test_content_list_converted_to_string(): + messages = [{"role": "user", "content": [{"type": "text", "text": "a"}, {"type": "text", "text": "b"}]}] + out = SambanovaConfig()._transform_messages(messages, model="MiniMax-M3") + assert isinstance(out[0]["content"], str) + assert "a" in out[0]["content"] and "b" in out[0]["content"]