chore: initial snapshot for gitea/github upload
This commit is contained in:
Your Name
@@ -0,0 +1,12 @@
|
||||
"""
|
||||
Pass-Through Endpoint Guardrail Translation
|
||||
|
||||
This module exists here (under litellm/llms/) so it can be auto-discovered by
|
||||
load_guardrail_translation_mappings() which scans for guardrail_translation
|
||||
directories under litellm/llms/.
|
||||
|
||||
The main passthrough endpoint implementation is in:
|
||||
litellm/proxy/pass_through_endpoints/
|
||||
|
||||
See guardrail_translation/README.md for more details.
|
||||
"""
|
||||
@@ -0,0 +1,41 @@
|
||||
# Pass-Through Endpoint Guardrail Translation
|
||||
|
||||
## Why This Exists Here
|
||||
|
||||
This module is located under `litellm/llms/` (instead of with the main passthrough code) because:
|
||||
|
||||
1. **Auto-discovery**: The `load_guardrail_translation_mappings()` function in `litellm/llms/__init__.py` scans for `guardrail_translation/` directories under `litellm/llms/`
|
||||
2. **Consistency**: All other guardrail translation handlers follow this pattern (e.g., `openai/chat/guardrail_translation/`, `anthropic/chat/guardrail_translation/`)
|
||||
|
||||
## Main Passthrough Implementation
|
||||
|
||||
The main passthrough endpoint implementation is in:
|
||||
|
||||
```
|
||||
litellm/proxy/pass_through_endpoints/
|
||||
├── pass_through_endpoints.py # Core passthrough routing logic
|
||||
├── passthrough_guardrails.py # Guardrail collection and field targeting
|
||||
├── jsonpath_extractor.py # JSONPath field extraction utility
|
||||
└── ...
|
||||
```
|
||||
|
||||
## What This Handler Does
|
||||
|
||||
The `PassThroughEndpointHandler` enables guardrails to run on passthrough endpoint requests by:
|
||||
|
||||
1. **Field Targeting**: Extracts specific fields from the request/response using JSONPath expressions configured in `request_fields` / `response_fields`
|
||||
2. **Full Payload Fallback**: If no field targeting is configured, processes the entire payload
|
||||
3. **Config Access**: Uses `get_passthrough_guardrails_config()` / `set_passthrough_guardrails_config()` helpers to access the passthrough guardrails configuration stored in request metadata
|
||||
|
||||
## Example Config
|
||||
|
||||
```yaml
|
||||
passthrough_endpoints:
|
||||
- path: "/v1/rerank"
|
||||
target: "https://api.cohere.com/v1/rerank"
|
||||
guardrails:
|
||||
bedrock-pre-guard:
|
||||
request_fields: ["query", "documents[*].text"]
|
||||
response_fields: ["results[*].text"]
|
||||
```
|
||||
|
||||
@@ -0,0 +1,15 @@
|
||||
"""Pass-Through Endpoint guardrail translation handler."""
|
||||
|
||||
from litellm.llms.pass_through.guardrail_translation.handler import (
|
||||
PassThroughEndpointHandler,
|
||||
)
|
||||
from litellm.types.utils import CallTypes
|
||||
|
||||
guardrail_translation_mappings = {
|
||||
CallTypes.pass_through: PassThroughEndpointHandler,
|
||||
}
|
||||
|
||||
__all__ = [
|
||||
"guardrail_translation_mappings",
|
||||
"PassThroughEndpointHandler",
|
||||
]
|
||||
@@ -0,0 +1,199 @@
|
||||
"""
|
||||
Pass-Through Endpoint Message Handler for Unified Guardrails
|
||||
|
||||
This module provides a handler for passthrough endpoint requests.
|
||||
It uses the field targeting configuration from litellm_logging_obj
|
||||
to extract specific fields for guardrail processing.
|
||||
"""
|
||||
|
||||
from typing import TYPE_CHECKING, Any, List, Optional
|
||||
|
||||
from litellm._logging import verbose_proxy_logger
|
||||
from litellm.llms.base_llm.guardrail_translation.base_translation import BaseTranslation
|
||||
from litellm.proxy._types import PassThroughGuardrailSettings
|
||||
from litellm.types.utils import GenericGuardrailAPIInputs
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from litellm.integrations.custom_guardrail import CustomGuardrail
|
||||
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
|
||||
|
||||
|
||||
class PassThroughEndpointHandler(BaseTranslation):
|
||||
"""
|
||||
Handler for processing passthrough endpoint requests with guardrails.
|
||||
|
||||
Uses passthrough_guardrails_config from litellm_logging_obj
|
||||
to determine which fields to extract for guardrail processing.
|
||||
"""
|
||||
|
||||
def _get_guardrail_settings(
|
||||
self,
|
||||
litellm_logging_obj: Optional["LiteLLMLoggingObj"],
|
||||
guardrail_name: Optional[str],
|
||||
) -> Optional[PassThroughGuardrailSettings]:
|
||||
"""
|
||||
Get the guardrail settings for a specific guardrail from logging_obj.
|
||||
"""
|
||||
from litellm.proxy.pass_through_endpoints.passthrough_guardrails import (
|
||||
PassthroughGuardrailHandler,
|
||||
)
|
||||
|
||||
if litellm_logging_obj is None:
|
||||
return None
|
||||
|
||||
passthrough_config = getattr(
|
||||
litellm_logging_obj, "passthrough_guardrails_config", None
|
||||
)
|
||||
if not passthrough_config or not guardrail_name:
|
||||
return None
|
||||
|
||||
return PassthroughGuardrailHandler.get_settings(
|
||||
passthrough_config, guardrail_name
|
||||
)
|
||||
|
||||
def _extract_text_for_guardrail(
|
||||
self,
|
||||
data: dict,
|
||||
field_expressions: Optional[List[str]],
|
||||
) -> str:
|
||||
"""
|
||||
Extract text from data for guardrail processing.
|
||||
|
||||
If field_expressions provided, extracts only those fields.
|
||||
Otherwise, returns the full payload as JSON.
|
||||
"""
|
||||
from litellm.proxy.pass_through_endpoints.jsonpath_extractor import (
|
||||
JsonPathExtractor,
|
||||
)
|
||||
|
||||
if field_expressions:
|
||||
text = JsonPathExtractor.extract_fields(
|
||||
data=data,
|
||||
jsonpath_expressions=field_expressions,
|
||||
)
|
||||
verbose_proxy_logger.debug(
|
||||
"PassThroughEndpointHandler: Extracted targeted fields: %s",
|
||||
text[:200] if text else None,
|
||||
)
|
||||
return text
|
||||
|
||||
# Use entire payload, excluding internal fields
|
||||
from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
|
||||
|
||||
payload_to_check = {
|
||||
k: v
|
||||
for k, v in data.items()
|
||||
if not k.startswith("_") and k not in ("metadata", "litellm_logging_obj")
|
||||
}
|
||||
verbose_proxy_logger.debug(
|
||||
"PassThroughEndpointHandler: Using full payload for guardrail"
|
||||
)
|
||||
return safe_dumps(payload_to_check)
|
||||
|
||||
async def process_input_messages(
|
||||
self,
|
||||
data: dict,
|
||||
guardrail_to_apply: "CustomGuardrail",
|
||||
litellm_logging_obj: Optional["LiteLLMLoggingObj"] = None,
|
||||
) -> Any:
|
||||
"""
|
||||
Process input by applying guardrails to targeted fields or full payload.
|
||||
"""
|
||||
guardrail_name = guardrail_to_apply.guardrail_name
|
||||
verbose_proxy_logger.debug(
|
||||
"PassThroughEndpointHandler: Processing input for guardrail=%s",
|
||||
guardrail_name,
|
||||
)
|
||||
|
||||
# Get field targeting settings for this guardrail
|
||||
settings = self._get_guardrail_settings(litellm_logging_obj, guardrail_name)
|
||||
field_expressions = settings.request_fields if settings else None
|
||||
|
||||
# Extract text to check
|
||||
text_to_check = self._extract_text_for_guardrail(data, field_expressions)
|
||||
|
||||
if not text_to_check:
|
||||
verbose_proxy_logger.debug(
|
||||
"PassThroughEndpointHandler: No text to check, skipping guardrail"
|
||||
)
|
||||
return data
|
||||
|
||||
# Apply guardrail (pass-through doesn't modify the text, just checks it)
|
||||
inputs = GenericGuardrailAPIInputs(texts=[text_to_check])
|
||||
# Include model information if available
|
||||
model = data.get("model")
|
||||
if model:
|
||||
inputs["model"] = model
|
||||
_guardrailed_inputs = await guardrail_to_apply.apply_guardrail(
|
||||
inputs=inputs,
|
||||
request_data=data,
|
||||
input_type="request",
|
||||
logging_obj=litellm_logging_obj,
|
||||
)
|
||||
|
||||
return data
|
||||
|
||||
async def process_output_response(
|
||||
self,
|
||||
response: Any,
|
||||
guardrail_to_apply: "CustomGuardrail",
|
||||
litellm_logging_obj: Optional["LiteLLMLoggingObj"] = None,
|
||||
user_api_key_dict: Optional[Any] = None,
|
||||
) -> Any:
|
||||
"""
|
||||
Process output response by applying guardrails to targeted fields.
|
||||
|
||||
Args:
|
||||
response: The response to process
|
||||
guardrail_to_apply: The guardrail instance to apply
|
||||
litellm_logging_obj: Optional logging object
|
||||
user_api_key_dict: User API key metadata to pass to guardrails
|
||||
"""
|
||||
if not isinstance(response, dict):
|
||||
verbose_proxy_logger.debug(
|
||||
"PassThroughEndpointHandler: Response is not a dict, skipping"
|
||||
)
|
||||
return response
|
||||
|
||||
guardrail_name = guardrail_to_apply.guardrail_name
|
||||
verbose_proxy_logger.debug(
|
||||
"PassThroughEndpointHandler: Processing output for guardrail=%s",
|
||||
guardrail_name,
|
||||
)
|
||||
|
||||
# Get field targeting settings for this guardrail
|
||||
settings = self._get_guardrail_settings(litellm_logging_obj, guardrail_name)
|
||||
field_expressions = settings.response_fields if settings else None
|
||||
|
||||
# Extract text to check
|
||||
text_to_check = self._extract_text_for_guardrail(response, field_expressions)
|
||||
|
||||
if not text_to_check:
|
||||
return response
|
||||
|
||||
# Create a request_data dict with response info and user API key metadata
|
||||
request_data: dict = (
|
||||
{"response": response}
|
||||
if not isinstance(response, dict)
|
||||
else response.copy()
|
||||
)
|
||||
|
||||
# Add user API key metadata with prefixed keys
|
||||
user_metadata = self.transform_user_api_key_dict_to_metadata(user_api_key_dict)
|
||||
if user_metadata:
|
||||
request_data["litellm_metadata"] = user_metadata
|
||||
|
||||
# Apply guardrail (pass-through doesn't modify the text, just checks it)
|
||||
inputs = GenericGuardrailAPIInputs(texts=[text_to_check])
|
||||
# Include model information from the response if available
|
||||
response_model = response.get("model") if isinstance(response, dict) else None
|
||||
if response_model:
|
||||
inputs["model"] = response_model
|
||||
_guardrailed_inputs = await guardrail_to_apply.apply_guardrail(
|
||||
inputs=inputs,
|
||||
request_data=request_data,
|
||||
input_type="response",
|
||||
logging_obj=litellm_logging_obj,
|
||||
)
|
||||
|
||||
return response
|
||||
Reference in New Issue
Block a user