Litellm dev 07 03 2025 p2 (#12301)

* fix(langfuse_prompt_management.py): add langfuse prompt version param support

Adds prompt versioning support

* test(test_langfuse_prompt_management.py): add unit test

* fix: fix linting errors

* fix: add prompt_version instrumentation

* fix(langfuse_otel.py): correctly read env var

* docs(langfuse_integration.md): clarify v3 recommendation is to use langfuse_otel

Closes https://github.com/BerriAI/litellm/issues/11500

* docs(langfuse_otel_integration.md): cleanup docs

* test: fix test

* test: update tests

* fix: fix linting error

docs/my-website/docs/observability/langfuse_integration.md

@@ -11,6 +11,13 @@ Example trace in Langfuse using multiple models via LiteLLM:
 <Image img={require('../../img/langfuse-example-trace-multiple-models-min.png')} />
 
 
+:::info
+
+For Langfuse v3, we recommend using the [Langfuse OTEL](./langfuse_otel_integration) integration.
+
+:::
+
+
 ## Usage with LiteLLM Proxy (LLM Gateway)
 
 👉 [**Follow this link to start sending logs to langfuse with LiteLLM Proxy server**](../proxy/logging)

docs/my-website/docs/observability/langfuse_otel_integration.md

@@ -1,7 +1,14 @@
-# Langfuse OpenTelemetry Integration
+import Tabs from '@theme/Tabs';
+import TabItem from '@theme/TabItem';
+
+import Image from '@theme/IdealImage';
+
+# 🪢 Langfuse OpenTelemetry Integration
 
 The Langfuse OpenTelemetry integration allows you to send LiteLLM traces and observability data to Langfuse using the OpenTelemetry protocol. This provides a standardized way to collect and analyze your LLM usage data.
 
+<Image img={require('../../img/langfuse_otel.png')} />
+
 ## Features
 
 - Automatic trace collection for all LiteLLM requests
@@ -108,15 +115,26 @@ litellm.callbacks = ["langfuse_otel"]
 
 Add the integration to your proxy configuration:
 
+1. Add the credentials to your environment variables
+
+```bash
+export LANGFUSE_PUBLIC_KEY="pk-lf-..."
+export LANGFUSE_SECRET_KEY="sk-lf-..."
+export LANGFUSE_HOST="https://us.cloud.langfuse.com"  # Default US region
+```
+
+2. Setup config.yaml
+
 ```yaml
 # config.yaml
 litellm_settings:
   callbacks: ["langfuse_otel"]
+```
 
-environment_variables:
-  LANGFUSE_PUBLIC_KEY: "pk-lf-..."
-  LANGFUSE_SECRET_KEY: "sk-lf-..."
-  LANGFUSE_HOST: "https://us.cloud.langfuse.com"  # Default US region
+3. Run the proxy
+
+```bash
+litellm --config /path/to/config.yaml
 ```
 
 ## Data Collected
@@ -163,11 +181,24 @@ This is automatically handled by the integration - you just need to provide the
 
 Enable verbose logging to see detailed information:
 
+<Tabs>
+<TabItem value="sdk" label="SDK">
+
 ```python
 import litellm
-litellm.set_verbose = True
+litellm._turn_on_debug()
 ```
 
+</TabItem>
+<TabItem value="proxy" label="PROXY">
+
+```bash
+export LITELLM_LOG="DEBUG"
+```
+
+</TabItem>
+</Tabs>
+
 This will show:
 - Endpoint resolution logic
 - Authentication header creation

docs/my-website/docs/proxy/prompt_management.md

@@ -210,6 +210,7 @@ These are the params you can pass to the `litellm.completion` function in SDK an
 ```
 prompt_id: str # required
 prompt_variables: Optional[dict] # optional
+prompt_version: Optional[int] # optional
 langfuse_public_key: Optional[str] # optional
 langfuse_secret: Optional[str] # optional
 langfuse_secret_key: Optional[str] # optional

litellm/integrations/anthropic_cache_control_hook.py

@@ -29,6 +29,7 @@ class AnthropicCacheControlHook(CustomPromptManagement):
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Apply cache control directives based on specified injection points.
@@ -80,10 +81,10 @@ class AnthropicCacheControlHook(CustomPromptManagement):
         # Case 1: Target by specific index
         if targetted_index is not None:
             if 0 <= targetted_index < len(messages):
-                messages[
-                    targetted_index
-                ] = AnthropicCacheControlHook._safe_insert_cache_control_in_message(
-                    messages[targetted_index], control
+                messages[targetted_index] = (
+                    AnthropicCacheControlHook._safe_insert_cache_control_in_message(
+                        messages[targetted_index], control
+                    )
                 )
         # Case 2: Target by role
         elif targetted_role is not None:

litellm/integrations/custom_logger.py

@@ -89,6 +89,7 @@ class CustomLogger:  # https://docs.litellm.ai/docs/observability/custom_callbac
         litellm_logging_obj: LiteLLMLoggingObj,
         tools: Optional[List[Dict]] = None,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Returns:
@@ -107,6 +108,7 @@ class CustomLogger:  # https://docs.litellm.ai/docs/observability/custom_callbac
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Returns:
@@ -408,9 +410,10 @@ class CustomLogger:  # https://docs.litellm.ai/docs/observability/custom_callbac
             if len(text) > max_length
             else text
         )
-    
 
-    def _select_metadata_field(self, request_kwargs: Optional[Dict] = None) -> Optional[str]:
+    def _select_metadata_field(
+        self, request_kwargs: Optional[Dict] = None
+    ) -> Optional[str]:
         """
         Select the metadata field to use for logging
 
@@ -418,9 +421,9 @@ class CustomLogger:  # https://docs.litellm.ai/docs/observability/custom_callbac
         2. Otherwise, use `metadata`
         """
         from litellm.constants import LITELLM_METADATA_FIELD, OLD_LITELLM_METADATA_FIELD
+
         if request_kwargs is None:
             return None
         if LITELLM_METADATA_FIELD in request_kwargs:
             return LITELLM_METADATA_FIELD
         return OLD_LITELLM_METADATA_FIELD
-        

litellm/integrations/custom_prompt_management.py

@@ -19,6 +19,7 @@ class CustomPromptManagement(CustomLogger, PromptManagementBase):
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Returns:
@@ -45,6 +46,7 @@ class CustomPromptManagement(CustomLogger, PromptManagementBase):
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> PromptManagementClient:
         raise NotImplementedError(
             "Custom prompt management does not support compile prompt helper"

litellm/integrations/humanloop.py

@@ -156,7 +156,12 @@ class HumanloopLogger(CustomLogger):
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
-    ) -> Tuple[str, List[AllMessageValues], dict,]:
+        prompt_version: Optional[int] = None,
+    ) -> Tuple[
+        str,
+        List[AllMessageValues],
+        dict,
+    ]:
         humanloop_api_key = dynamic_callback_params.get(
             "humanloop_api_key"
         ) or get_secret_str("HUMANLOOP_API_KEY")

litellm/integrations/langfuse/langfuse_otel.py

@@ -1,6 +1,7 @@
 import base64
 import os
 from typing import TYPE_CHECKING, Any, Union
+from urllib.parse import quote
 
 from litellm._logging import verbose_logger
 from litellm.integrations.arize import _utils
@@ -9,10 +10,11 @@ from litellm.types.integrations.langfuse_otel import LangfuseOtelConfig
 if TYPE_CHECKING:
     from opentelemetry.trace import Span as _Span
 
+    from litellm.integrations.opentelemetry import (
+        OpenTelemetryConfig as _OpenTelemetryConfig,
+    )
     from litellm.types.integrations.arize import Protocol as _Protocol
 
-    from litellm.integrations.opentelemetry import OpenTelemetryConfig as _OpenTelemetryConfig
-
     Protocol = _Protocol
     OpenTelemetryConfig = _OpenTelemetryConfig
     Span = Union[_Span, Any]
@@ -54,7 +56,7 @@ class LangfuseOtelLogger:
         """
         public_key = os.environ.get("LANGFUSE_PUBLIC_KEY", None)
         secret_key = os.environ.get("LANGFUSE_SECRET_KEY", None)
-        
+
         if not public_key or not secret_key:
             raise ValueError(
                 "LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY must be set for Langfuse OpenTelemetry integration."
@@ -62,7 +64,7 @@ class LangfuseOtelLogger:
 
         # Determine endpoint - default to US cloud
         langfuse_host = os.environ.get("LANGFUSE_HOST", None)
-        
+
         if langfuse_host:
             # If LANGFUSE_HOST is provided, construct OTEL endpoint from it
             if not langfuse_host.startswith("http"):
@@ -77,13 +79,13 @@ class LangfuseOtelLogger:
         # Create Basic Auth header
         auth_string = f"{public_key}:{secret_key}"
         auth_header = base64.b64encode(auth_string.encode()).decode()
-        otlp_auth_headers = f"Authorization=Basic {auth_header}"
+        # URL encode the entire header value as required by OpenTelemetry specification
+        otlp_auth_headers = f"Authorization={quote(f'Basic {auth_header}')}"
 
         # Set standard OTEL environment variables
         os.environ["OTEL_EXPORTER_OTLP_ENDPOINT"] = endpoint
         os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = otlp_auth_headers
 
         return LangfuseOtelConfig(
-            otlp_auth_headers=otlp_auth_headers,
-            protocol="otlp_http"
-        ) 
+            otlp_auth_headers=otlp_auth_headers, protocol="otlp_http"
+        )

litellm/integrations/langfuse/langfuse_prompt_management.py

@@ -134,8 +134,14 @@ class LangfusePromptManagement(LangFuseLogger, PromptManagementBase, CustomLogge
         langfuse_prompt_id: str,
         langfuse_client: LangfuseClass,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> PROMPT_CLIENT:
-        return langfuse_client.get_prompt(langfuse_prompt_id, label=prompt_label)
+
+        prompt_client = langfuse_client.get_prompt(
+            langfuse_prompt_id, label=prompt_label, version=prompt_version
+        )
+
+        return prompt_client
 
     def _compile_prompt(
         self,
@@ -180,7 +186,12 @@ class LangfusePromptManagement(LangFuseLogger, PromptManagementBase, CustomLogge
         litellm_logging_obj: LiteLLMLoggingObj,
         tools: Optional[List[Dict]] = None,
         prompt_label: Optional[str] = None,
-    ) -> Tuple[str, List[AllMessageValues], dict,]:
+        prompt_version: Optional[int] = None,
+    ) -> Tuple[
+        str,
+        List[AllMessageValues],
+        dict,
+    ]:
         return self.get_chat_completion_prompt(
             model,
             messages,
@@ -189,6 +200,7 @@ class LangfusePromptManagement(LangFuseLogger, PromptManagementBase, CustomLogge
             prompt_variables,
             dynamic_callback_params,
             prompt_label=prompt_label,
+            prompt_version=prompt_version,
         )
 
     def should_run_prompt_management(
@@ -203,7 +215,8 @@ class LangfusePromptManagement(LangFuseLogger, PromptManagementBase, CustomLogge
             langfuse_host=dynamic_callback_params.get("langfuse_host"),
         )
         langfuse_prompt_client = self._get_prompt_from_id(
-            langfuse_prompt_id=prompt_id, langfuse_client=langfuse_client
+            langfuse_prompt_id=prompt_id,
+            langfuse_client=langfuse_client,
         )
         return langfuse_prompt_client is not None
 
@@ -213,6 +226,7 @@ class LangfusePromptManagement(LangFuseLogger, PromptManagementBase, CustomLogge
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> PromptManagementClient:
         langfuse_client = langfuse_client_init(
             langfuse_public_key=dynamic_callback_params.get("langfuse_public_key"),
@@ -224,6 +238,7 @@ class LangfusePromptManagement(LangFuseLogger, PromptManagementBase, CustomLogge
             langfuse_prompt_id=prompt_id,
             langfuse_client=langfuse_client,
             prompt_label=prompt_label,
+            prompt_version=prompt_version,
         )
 
         ## SET PROMPT

litellm/integrations/prompt_management_base.py

@@ -34,6 +34,7 @@ class PromptManagementBase(ABC):
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> PromptManagementClient:
         pass
 
@@ -51,12 +52,14 @@ class PromptManagementBase(ABC):
         client_messages: List[AllMessageValues],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> PromptManagementClient:
         compiled_prompt_client = self._compile_prompt_helper(
             prompt_id=prompt_id,
             prompt_variables=prompt_variables,
             dynamic_callback_params=dynamic_callback_params,
             prompt_label=prompt_label,
+            prompt_version=prompt_version,
         )
 
         try:
@@ -86,6 +89,7 @@ class PromptManagementBase(ABC):
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         if prompt_id is None:
             raise ValueError("prompt_id is required for Prompt Management Base class")
@@ -100,6 +104,7 @@ class PromptManagementBase(ABC):
             client_messages=messages,
             dynamic_callback_params=dynamic_callback_params,
             prompt_label=prompt_label,
+            prompt_version=prompt_version,
         )
 
         completed_messages = prompt_template["completed_messages"] or messages

litellm/integrations/vector_store_integrations/bedrock_vector_store.py

@@ -77,6 +77,7 @@ class BedrockVectorStore(BaseVectorStore, BaseAWSLLM):
         litellm_logging_obj: LiteLLMLoggingObj,
         tools: Optional[List[Dict]] = None,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Retrieves the context from the Bedrock Knowledge Base and appends it to the messages.
@@ -129,9 +130,9 @@ class BedrockVectorStore(BaseVectorStore, BaseAWSLLM):
                     )
                 )
 
-            litellm_logging_obj.model_call_details[
-                "vector_store_request_metadata"
-            ] = vector_store_request_metadata
+            litellm_logging_obj.model_call_details["vector_store_request_metadata"] = (
+                vector_store_request_metadata
+            )
 
         return model, messages, non_default_params
 
@@ -143,9 +144,9 @@ class BedrockVectorStore(BaseVectorStore, BaseAWSLLM):
         """
         Transform a BedrockKBResponse to a VectorStoreSearchResponse
         """
-        retrieval_results: Optional[
-            List[BedrockKBRetrievalResult]
-        ] = bedrock_kb_response.get("retrievalResults", None)
+        retrieval_results: Optional[List[BedrockKBRetrievalResult]] = (
+            bedrock_kb_response.get("retrievalResults", None)
+        )
         vector_store_search_response: VectorStoreSearchResponse = (
             VectorStoreSearchResponse(search_query=query, data=[])
         )

litellm/integrations/vector_stores/bedrock_vector_store.py

@@ -77,6 +77,7 @@ class BedrockVectorStore(BaseVectorStore, BaseAWSLLM):
         litellm_logging_obj: LiteLLMLoggingObj,
         tools: Optional[List[Dict]] = None,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Retrieves the context from the Bedrock Knowledge Base and appends it to the messages.
@@ -129,9 +130,9 @@ class BedrockVectorStore(BaseVectorStore, BaseAWSLLM):
                     )
                 )
 
-            litellm_logging_obj.model_call_details[
-                "vector_store_request_metadata"
-            ] = vector_store_request_metadata
+            litellm_logging_obj.model_call_details["vector_store_request_metadata"] = (
+                vector_store_request_metadata
+            )
 
         return model, messages, non_default_params
 
@@ -143,9 +144,9 @@ class BedrockVectorStore(BaseVectorStore, BaseAWSLLM):
         """
         Transform a BedrockKBResponse to a VectorStoreSearchResponse
         """
-        retrieval_results: Optional[
-            List[BedrockKBRetrievalResult]
-        ] = bedrock_kb_response.get("retrievalResults", None)
+        retrieval_results: Optional[List[BedrockKBRetrievalResult]] = (
+            bedrock_kb_response.get("retrievalResults", None)
+        )
         vector_store_search_response: VectorStoreSearchResponse = (
             VectorStoreSearchResponse(search_query=query, data=[])
         )

litellm/litellm_core_utils/initialize_dynamic_callback_params.py

@@ -18,6 +18,7 @@ def initialize_standard_callback_dynamic_params(
         _supported_callback_params = (
             StandardCallbackDynamicParams.__annotations__.keys()
         )
+
         for param in _supported_callback_params:
             if param in kwargs:
                 _param_value = kwargs.pop(param)

litellm/litellm_core_utils/litellm_logging.py

@@ -435,6 +435,7 @@ class Logging(LiteLLMLoggingBaseClass):
 
         checks if langfuse_secret_key, gcs_bucket_name in kwargs and sets the corresponding attributes in StandardCallbackDynamicParams
         """
+
         return _initialize_standard_callback_dynamic_params(kwargs)
 
     def initialize_standard_built_in_tools_params(
@@ -556,6 +557,7 @@ class Logging(LiteLLMLoggingBaseClass):
         prompt_variables: Optional[dict],
         prompt_management_logger: Optional[CustomLogger] = None,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         custom_logger = (
             prompt_management_logger
@@ -577,6 +579,7 @@ class Logging(LiteLLMLoggingBaseClass):
                 prompt_variables=prompt_variables,
                 dynamic_callback_params=self.standard_callback_dynamic_params,
                 prompt_label=prompt_label,
+                prompt_version=prompt_version,
             )
         self.messages = messages
         return model, messages, non_default_params
@@ -591,6 +594,7 @@ class Logging(LiteLLMLoggingBaseClass):
         prompt_management_logger: Optional[CustomLogger] = None,
         tools: Optional[List[Dict]] = None,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         custom_logger = (
             prompt_management_logger
@@ -614,6 +618,7 @@ class Logging(LiteLLMLoggingBaseClass):
                 litellm_logging_obj=self,
                 tools=tools,
                 prompt_label=prompt_label,
+                prompt_version=prompt_version,
             )
         self.messages = messages
         return model, messages, non_default_params
@@ -2953,8 +2958,6 @@ def set_callbacks(callback_list, function_id=None):  # noqa: PLR0915
                     import sentry_sdk
                 from sentry_sdk.scrubber import EventScrubber
 
-                
-
                 sentry_sdk_instance = sentry_sdk
                 sentry_trace_rate = (
                     os.environ.get("SENTRY_API_TRACE_RATE")
@@ -2974,8 +2977,7 @@ def set_callbacks(callback_list, function_id=None):  # noqa: PLR0915
                     ),
                     send_default_pii=False,  # Prevent sending Personal Identifiable Information
                     event_scrubber=EventScrubber(
-                        denylist=SENTRY_DENYLIST,
-                        pii_denylist=SENTRY_PII_DENYLIST
+                        denylist=SENTRY_DENYLIST, pii_denylist=SENTRY_PII_DENYLIST
                     ),
                 )
                 capture_exception = sentry_sdk_instance.capture_exception
@@ -3033,6 +3035,7 @@ def set_callbacks(callback_list, function_id=None):  # noqa: PLR0915
                 s3Logger = S3Logger()
             elif callback == "wandb":
                 from litellm.integrations.weights_biases import WeightsBiasesLogger
+
                 weightsBiasesLogger = WeightsBiasesLogger()
             elif callback == "logfire":
                 logfireLogger = LogfireLogger()
@@ -3087,6 +3090,7 @@ def _init_custom_logger_compatible_class(  # noqa: PLR0915
             return _openmeter_logger  # type: ignore
         elif logging_integration == "braintrust":
             from litellm.integrations.braintrust_logging import BraintrustLogger
+
             for callback in _in_memory_loggers:
                 if isinstance(callback, BraintrustLogger):
                     return callback  # type: ignore
@@ -3453,6 +3457,7 @@ def get_custom_logger_compatible_class(  # noqa: PLR0915
                     return callback
         elif logging_integration == "braintrust":
             from litellm.integrations.braintrust_logging import BraintrustLogger
+
             for callback in _in_memory_loggers:
                 if isinstance(callback, BraintrustLogger):
                     return callback

litellm/main.py

@@ -438,6 +438,7 @@ async def acompletion(
             prompt_variables=kwargs.get("prompt_variables", None),
             tools=tools,
             prompt_label=kwargs.get("prompt_label", None),
+            prompt_version=kwargs.get("prompt_version", None),
         )
         #########################################################
         # if the chat completion logging hook removed all tools,
@@ -1062,6 +1063,7 @@ def completion(  # type: ignore # noqa: PLR0915
             prompt_id=prompt_id,
             prompt_variables=prompt_variables,
             prompt_label=kwargs.get("prompt_label", None),
+            prompt_version=kwargs.get("prompt_version", None),
         )
 
     try:
@@ -1950,7 +1952,7 @@ def completion(  # type: ignore # noqa: PLR0915
                 or get_secret("MISTRAL_API_BASE")
                 or "https://api.mistral.ai/v1"
             )
-            
+
             response = base_llm_http_handler.completion(
                 model=model,
                 messages=messages,
@@ -2615,9 +2617,7 @@ def completion(  # type: ignore # noqa: PLR0915
             api_base = api_base or litellm.api_base or get_secret("VERTEXAI_API_BASE")
 
             new_params = deepcopy(optional_params)
-            if (
-                vertex_partner_models_chat_completion.is_vertex_partner_model(model)
-            ):
+            if vertex_partner_models_chat_completion.is_vertex_partner_model(model):
                 model_response = vertex_partner_models_chat_completion.completion(
                     model=model,
                     messages=messages,

litellm/proxy/_new_secret_config.yaml

@@ -3,11 +3,11 @@ model_list:
     litellm_params:
       model: openai/gpt-3.5-turbo
       api_key: os.environ/OPENAI_API_KEY
-
-guardrails:
-  - guardrail_name: azure-prompt-shield
+  - model_name: langfuse-model
     litellm_params:
-      guardrail: azure/prompt_shield
-      mode: pre_call # "During_call" is also available
-      api_key: os.environ/AZURE_GUARDRAIL_API_KEY
-      api_base: os.environ/AZURE_GUARDRAIL_API_BASE 
+      model: langfuse/langfuse-model
+      prompt_id: test-chat-prompt 
+      prompt_version: 4
+
+litellm_settings:
+  callbacks: ["langfuse_otel"]

litellm/proxy/custom_prompt_management.py

@@ -16,6 +16,7 @@ class X42PromptManagement(CustomPromptManagement):
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
         prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         """
         Returns:

litellm/router.py

@@ -1735,6 +1735,7 @@ class Router:
         kwargs: Dict[str, Any],
     ):
         litellm_logging_object = kwargs.get("litellm_logging_obj", None)
+
         if litellm_logging_object is None:
             litellm_logging_object, kwargs = function_setup(
                 **{
@@ -1770,6 +1771,10 @@ class Router:
             "litellm_params"
         ].get("prompt_label", None)
 
+        prompt_version = kwargs.get(
+            "prompt_version", None
+        ) or prompt_management_deployment["litellm_params"].get("prompt_version", None)
+
         if prompt_id is None or not isinstance(prompt_id, str):
             raise ValueError(
                 f"Prompt ID is not set or not a string. Got={prompt_id}, type={type(prompt_id)}"
@@ -1790,6 +1795,7 @@ class Router:
             prompt_id=prompt_id,
             prompt_variables=prompt_variables,
             prompt_label=prompt_label,
+            prompt_version=prompt_version,
         )
 
         kwargs = {**data, **kwargs, **optional_params}

litellm/types/utils.py

@@ -2072,6 +2072,9 @@ class StandardCallbackDynamicParams(TypedDict, total=False):
     langfuse_secret_key: Optional[str]
     langfuse_host: Optional[str]
 
+    # Langfuse prompt version
+    langfuse_prompt_version: Optional[int]
+
     # GCS dynamic params
     gcs_bucket_name: Optional[str]
     gcs_path_service_account: Optional[str]
@@ -2112,6 +2115,7 @@ all_litellm_params = [
     "prompt_id",
     "provider_specific_header",
     "prompt_variables",
+    "prompt_version",
     "api_base",
     "force_timeout",
     "logger_fn",

tests/test_litellm/integrations/langfuse/test_langfuse_prompt_management.py

@@ -0,0 +1,32 @@
+import os
+from unittest.mock import patch
+
+from litellm.integrations.langfuse.langfuse_prompt_management import (
+    LangfusePromptManagement,
+)
+
+
+class TestLangfusePromptManagement:
+    def test_get_prompt_from_id(self):
+        langfuse_prompt_management = LangfusePromptManagement()
+        with patch.object(
+            langfuse_prompt_management, "should_run_prompt_management"
+        ) as mock_should_run_prompt_management, patch.object(
+            langfuse_prompt_management, "_get_prompt_from_id"
+        ) as mock_get_prompt_from_id:
+            mock_should_run_prompt_management.return_value = True
+            chat_completion_prompt = (
+                langfuse_prompt_management.get_chat_completion_prompt(
+                    model="langfuse/langfuse-model",
+                    messages=[{"role": "user", "content": "Hello, how are you?"}],
+                    non_default_params={},
+                    prompt_id="test-chat-prompt",
+                    prompt_variables={},
+                    dynamic_callback_params={},
+                    prompt_version=4,
+                )
+            )
+
+            mock_get_prompt_from_id.assert_called_once()
+            print(mock_get_prompt_from_id.call_args.kwargs)
+            assert mock_get_prompt_from_id.call_args.kwargs["prompt_version"] == 4

tests/test_litellm/integrations/test_custom_prompt_management.py

@@ -33,7 +33,8 @@ class TestCustomPromptManagement(CustomPromptManagement):
         prompt_id: Optional[str],
         prompt_variables: Optional[dict],
         dynamic_callback_params: StandardCallbackDynamicParams,
-        prompt_label: Optional[str],
+        prompt_label: Optional[str] = None,
+        prompt_version: Optional[int] = None,
     ) -> Tuple[str, List[AllMessageValues], dict]:
         print(
             "TestCustomPromptManagement: running get_chat_completion_prompt for prompt_id: ",