You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
langchain/libs/community/langchain_community/callbacks/whylabs_callback.py

188 lines
7.7 KiB
Python

from __future__ import annotations
import logging
from typing import TYPE_CHECKING, Any, Optional
from langchain_core.callbacks import BaseCallbackHandler
from langchain_core.utils import get_from_env, guard_import
if TYPE_CHECKING:
from whylogs.api.logger.logger import Logger
diagnostic_logger = logging.getLogger(__name__)
def import_langkit(
sentiment: bool = False,
toxicity: bool = False,
themes: bool = False,
) -> Any:
"""Import the langkit python package and raise an error if it is not installed.
Args:
sentiment: Whether to import the langkit.sentiment module. Defaults to False.
toxicity: Whether to import the langkit.toxicity module. Defaults to False.
themes: Whether to import the langkit.themes module. Defaults to False.
Returns:
The imported langkit module.
"""
langkit = guard_import("langkit")
guard_import("langkit.regexes")
guard_import("langkit.textstat")
if sentiment:
guard_import("langkit.sentiment")
if toxicity:
guard_import("langkit.toxicity")
if themes:
guard_import("langkit.themes")
return langkit
class WhyLabsCallbackHandler(BaseCallbackHandler):
"""
Callback Handler for logging to WhyLabs. This callback handler utilizes
`langkit` to extract features from the prompts & responses when interacting with
an LLM. These features can be used to guardrail, evaluate, and observe interactions
over time to detect issues relating to hallucinations, prompt engineering,
or output validation. LangKit is an LLM monitoring toolkit developed by WhyLabs.
Here are some examples of what can be monitored with LangKit:
* Text Quality
- readability score
- complexity and grade scores
* Text Relevance
- Similarity scores between prompt/responses
- Similarity scores against user-defined themes
- Topic classification
* Security and Privacy
- patterns - count of strings matching a user-defined regex pattern group
- jailbreaks - similarity scores with respect to known jailbreak attempts
- prompt injection - similarity scores with respect to known prompt attacks
- refusals - similarity scores with respect to known LLM refusal responses
* Sentiment and Toxicity
- sentiment analysis
- toxicity analysis
For more information, see https://docs.whylabs.ai/docs/language-model-monitoring
or check out the LangKit repo here: https://github.com/whylabs/langkit
---
Args:
api_key (Optional[str]): WhyLabs API key. Optional because the preferred
way to specify the API key is with environment variable
WHYLABS_API_KEY.
org_id (Optional[str]): WhyLabs organization id to write profiles to.
Optional because the preferred way to specify the organization id is
with environment variable WHYLABS_DEFAULT_ORG_ID.
dataset_id (Optional[str]): WhyLabs dataset id to write profiles to.
Optional because the preferred way to specify the dataset id is
with environment variable WHYLABS_DEFAULT_DATASET_ID.
sentiment (bool): Whether to enable sentiment analysis. Defaults to False.
toxicity (bool): Whether to enable toxicity analysis. Defaults to False.
themes (bool): Whether to enable theme analysis. Defaults to False.
"""
def __init__(self, logger: Logger, handler: Any):
"""Initiate the rolling logger."""
super().__init__()
if hasattr(handler, "init"):
handler.init(self)
if hasattr(handler, "_get_callbacks"):
self._callbacks = handler._get_callbacks()
else:
self._callbacks = dict()
diagnostic_logger.warning("initialized handler without callbacks.")
self._logger = logger
def flush(self) -> None:
"""Explicitly write current profile if using a rolling logger."""
if self._logger and hasattr(self._logger, "_do_rollover"):
self._logger._do_rollover()
diagnostic_logger.info("Flushing WhyLabs logger, writing profile...")
def close(self) -> None:
"""Close any loggers to allow writing out of any profiles before exiting."""
if self._logger and hasattr(self._logger, "close"):
self._logger.close()
diagnostic_logger.info("Closing WhyLabs logger, see you next time!")
def __enter__(self) -> WhyLabsCallbackHandler:
return self
def __exit__(
self, exception_type: Any, exception_value: Any, traceback: Any
) -> None:
self.close()
@classmethod
def from_params(
cls,
*,
api_key: Optional[str] = None,
org_id: Optional[str] = None,
dataset_id: Optional[str] = None,
sentiment: bool = False,
toxicity: bool = False,
themes: bool = False,
logger: Optional[Logger] = None,
) -> WhyLabsCallbackHandler:
"""Instantiate whylogs Logger from params.
Args:
api_key (Optional[str]): WhyLabs API key. Optional because the preferred
way to specify the API key is with environment variable
WHYLABS_API_KEY.
org_id (Optional[str]): WhyLabs organization id to write profiles to.
If not set must be specified in environment variable
WHYLABS_DEFAULT_ORG_ID.
dataset_id (Optional[str]): The model or dataset this callback is gathering
telemetry for. If not set must be specified in environment variable
WHYLABS_DEFAULT_DATASET_ID.
sentiment (bool): If True will initialize a model to perform
sentiment analysis compound score. Defaults to False and will not gather
this metric.
toxicity (bool): If True will initialize a model to score
toxicity. Defaults to False and will not gather this metric.
themes (bool): If True will initialize a model to calculate
distance to configured themes. Defaults to None and will not gather this
metric.
logger (Optional[Logger]): If specified will bind the configured logger as
the telemetry gathering agent. Defaults to LangKit schema with periodic
WhyLabs writer.
"""
# langkit library will import necessary whylogs libraries
import_langkit(sentiment=sentiment, toxicity=toxicity, themes=themes)
why = guard_import("whylogs")
get_callback_instance = guard_import(
"langkit.callback_handler"
).get_callback_instance
WhyLabsWriter = guard_import("whylogs.api.writer.whylabs").WhyLabsWriter
udf_schema = guard_import("whylogs.experimental.core.udf_schema").udf_schema
if logger is None:
api_key = api_key or get_from_env("api_key", "WHYLABS_API_KEY")
org_id = org_id or get_from_env("org_id", "WHYLABS_DEFAULT_ORG_ID")
dataset_id = dataset_id or get_from_env(
"dataset_id", "WHYLABS_DEFAULT_DATASET_ID"
)
whylabs_writer = WhyLabsWriter(
api_key=api_key, org_id=org_id, dataset_id=dataset_id
)
whylabs_logger = why.logger(
mode="rolling", interval=5, when="M", schema=udf_schema()
)
whylabs_logger.append_writer(writer=whylabs_writer)
else:
diagnostic_logger.info("Using passed in whylogs logger {logger}")
whylabs_logger = logger
callback_handler_cls = get_callback_instance(logger=whylabs_logger, impl=cls)
diagnostic_logger.info(
"Started whylogs Logger with WhyLabsWriter and initialized LangKit. 📝"
)
return callback_handler_cls