phoenix/litellm-mirror
0
0
Fork 1
mirror of https://github.com/BerriAI/litellm.git synced 2025-04-25 02:34:29 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add /vllm/* and /mistral/* passthrough endpoints (adds support for Mistral OCR via passthrough)

Browse source 
* feat(llm_passthrough_endpoints.py): support mistral passthrough

Closes https://github.com/BerriAI/litellm/issues/9051

* feat(llm_passthrough_endpoints.py): initial commit for adding vllm passthrough route

* feat(vllm/common_utils.py): add new vllm model info route

make it possible to use vllm passthrough route via factory function

* fix(llm_passthrough_endpoints.py): add all methods to vllm passthrough route

* fix: fix linting error

* fix: fix linting error

* fix: fix ruff check

* fix(proxy/_types.py): add new passthrough routes

* docs(config_settings.md): add mistral env vars to docs
This commit is contained in:
Krish Dholakia 2025-04-14 22:06:33 -07:00 • committed by GitHub
parent 8faf56922c
commit 9b0f871129
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 450 additions and 176 deletions
Download patch file Download diff file Expand all files Collapse all files

53
litellm/utils.py
View file

@ -516,9 +516,9 @@ def function_setup( # noqa: PLR0915
function_id: Optional[str] = kwargs["id"] if "id" in kwargs else None
## DYNAMIC CALLBACKS ##
dynamic_callbacks: Optional[List[Union[str, Callable, CustomLogger]]] = (
kwargs.pop("callbacks", None)
)
dynamic_callbacks: Optional[
List[Union[str, Callable, CustomLogger]]
] = kwargs.pop("callbacks", None)
all_callbacks = get_dynamic_callbacks(dynamic_callbacks=dynamic_callbacks)
if len(all_callbacks) > 0:
@ -1202,9 +1202,9 @@ def client(original_function): # noqa: PLR0915
exception=e,
retry_policy=kwargs.get("retry_policy"),
)
kwargs["retry_policy"] = (
reset_retry_policy()
) # prevent infinite loops
kwargs[
"retry_policy"
] = reset_retry_policy() # prevent infinite loops
litellm.num_retries = (
None # set retries to None to prevent infinite loops
)
@ -3013,16 +3013,16 @@ def get_optional_params( # noqa: PLR0915
True # so that main.py adds the function call to the prompt
)
if "tools" in non_default_params:
optional_params["functions_unsupported_model"] = (
non_default_params.pop("tools")
)
optional_params[
"functions_unsupported_model"
] = non_default_params.pop("tools")
non_default_params.pop(
"tool_choice", None
) # causes ollama requests to hang
elif "functions" in non_default_params:
optional_params["functions_unsupported_model"] = (
non_default_params.pop("functions")
)
optional_params[
"functions_unsupported_model"
] = non_default_params.pop("functions")
elif (
litellm.add_function_to_prompt
): # if user opts to add it to prompt instead
@ -3045,10 +3045,10 @@ def get_optional_params( # noqa: PLR0915
if "response_format" in non_default_params:
if provider_config is not None:
non_default_params["response_format"] = (
provider_config.get_json_schema_from_pydantic_object(
response_format=non_default_params["response_format"]
)
non_default_params[
"response_format"
] = provider_config.get_json_schema_from_pydantic_object(
response_format=non_default_params["response_format"]
)
else:
non_default_params["response_format"] = type_to_response_format_param(
@ -4064,9 +4064,9 @@ def _count_characters(text: str) -> int:
def get_response_string(response_obj: Union[ModelResponse, ModelResponseStream]) -> str:
_choices: Union[List[Union[Choices, StreamingChoices]], List[StreamingChoices]] = (
response_obj.choices
)
_choices: Union[
List[Union[Choices, StreamingChoices]], List[StreamingChoices]
] = response_obj.choices
response_str = ""
for choice in _choices:
@ -4458,14 +4458,14 @@ def _get_model_info_helper( # noqa: PLR0915
if combined_model_name in litellm.model_cost:
key = combined_model_name
_model_info = _get_model_info_from_model_cost(key=key)
_model_info = _get_model_info_from_model_cost(key=cast(str, key))
if not _check_provider_match(
model_info=_model_info, custom_llm_provider=custom_llm_provider
):
_model_info = None
if _model_info is None and model in litellm.model_cost:
key = model
_model_info = _get_model_info_from_model_cost(key=key)
_model_info = _get_model_info_from_model_cost(key=cast(str, key))
if not _check_provider_match(
model_info=_model_info, custom_llm_provider=custom_llm_provider
):
@ -4475,21 +4475,21 @@ def _get_model_info_helper( # noqa: PLR0915
and combined_stripped_model_name in litellm.model_cost
):
key = combined_stripped_model_name
_model_info = _get_model_info_from_model_cost(key=key)
_model_info = _get_model_info_from_model_cost(key=cast(str, key))
if not _check_provider_match(
model_info=_model_info, custom_llm_provider=custom_llm_provider
):
_model_info = None
if _model_info is None and stripped_model_name in litellm.model_cost:
key = stripped_model_name
_model_info = _get_model_info_from_model_cost(key=key)
_model_info = _get_model_info_from_model_cost(key=cast(str, key))
if not _check_provider_match(
model_info=_model_info, custom_llm_provider=custom_llm_provider
):
_model_info = None
if _model_info is None and split_model in litellm.model_cost:
key = split_model
_model_info = _get_model_info_from_model_cost(key=key)
_model_info = _get_model_info_from_model_cost(key=cast(str, key))
if not _check_provider_match(
model_info=_model_info, custom_llm_provider=custom_llm_provider
):
@ -6510,7 +6510,12 @@ class ProviderConfigManager:
return litellm.AnthropicModelInfo()
elif LlmProviders.XAI == provider:
return litellm.XAIModelInfo()
elif LlmProviders.VLLM == provider:
from litellm.llms.vllm.common_utils import (
VLLMModelInfo, # experimental approach, to reduce bloat on __init__.py
)
return VLLMModelInfo()
return None
@staticmethod