langchain/libs/community/langchain_community/callbacks/confident_callback.py
Mohammad Mohtashim 43db4cd20e
core[major]: On Tool End Observation Casting Fix (#18798)
This PR updates the on_tool_end handlers to return the raw output from the tool instead of casting it to a string. 

This is technically a breaking change, though it's impact is expected to be somewhat minimal. It will fix behavior in `astream_events` as well.

Fixes the following issue #18760 raised by @eyurtsev

---------

Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
2024-03-11 10:59:04 -04:00

184 lines
6.2 KiB
Python

# flake8: noqa
import os
import warnings
from typing import Any, Dict, List, Optional, Union
from langchain_core.callbacks import BaseCallbackHandler
from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.outputs import LLMResult
class DeepEvalCallbackHandler(BaseCallbackHandler):
"""Callback Handler that logs into deepeval.
Args:
implementation_name: name of the `implementation` in deepeval
metrics: A list of metrics
Raises:
ImportError: if the `deepeval` package is not installed.
Examples:
>>> from langchain_community.llms import OpenAI
>>> from langchain_community.callbacks import DeepEvalCallbackHandler
>>> from deepeval.metrics import AnswerRelevancy
>>> metric = AnswerRelevancy(minimum_score=0.3)
>>> deepeval_callback = DeepEvalCallbackHandler(
... implementation_name="exampleImplementation",
... metrics=[metric],
... )
>>> llm = OpenAI(
... temperature=0,
... callbacks=[deepeval_callback],
... verbose=True,
... openai_api_key="API_KEY_HERE",
... )
>>> llm.generate([
... "What is the best evaluation tool out there? (no bias at all)",
... ])
"Deepeval, no doubt about it."
"""
REPO_URL: str = "https://github.com/confident-ai/deepeval"
ISSUES_URL: str = f"{REPO_URL}/issues"
BLOG_URL: str = "https://docs.confident-ai.com" # noqa: E501
def __init__(
self,
metrics: List[Any],
implementation_name: Optional[str] = None,
) -> None:
"""Initializes the `deepevalCallbackHandler`.
Args:
implementation_name: Name of the implementation you want.
metrics: What metrics do you want to track?
Raises:
ImportError: if the `deepeval` package is not installed.
ConnectionError: if the connection to deepeval fails.
"""
super().__init__()
# Import deepeval (not via `import_deepeval` to keep hints in IDEs)
try:
import deepeval # ignore: F401,I001
except ImportError:
raise ImportError(
"""To use the deepeval callback manager you need to have the
`deepeval` Python package installed. Please install it with
`pip install deepeval`"""
)
if os.path.exists(".deepeval"):
warnings.warn(
"""You are currently not logging anything to the dashboard, we
recommend using `deepeval login`."""
)
# Set the deepeval variables
self.implementation_name = implementation_name
self.metrics = metrics
warnings.warn(
(
"The `DeepEvalCallbackHandler` is currently in beta and is subject to"
" change based on updates to `langchain`. Please report any issues to"
f" {self.ISSUES_URL} as an `integration` issue."
),
)
def on_llm_start(
self, serialized: Dict[str, Any], prompts: List[str], **kwargs: Any
) -> None:
"""Store the prompts"""
self.prompts = prompts
def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
"""Do nothing when a new token is generated."""
pass
def on_llm_end(self, response: LLMResult, **kwargs: Any) -> None:
"""Log records to deepeval when an LLM ends."""
from deepeval.metrics.answer_relevancy import AnswerRelevancy
from deepeval.metrics.bias_classifier import UnBiasedMetric
from deepeval.metrics.metric import Metric
from deepeval.metrics.toxic_classifier import NonToxicMetric
for metric in self.metrics:
for i, generation in enumerate(response.generations):
# Here, we only measure the first generation's output
output = generation[0].text
query = self.prompts[i]
if isinstance(metric, AnswerRelevancy):
result = metric.measure(
output=output,
query=query,
)
print(f"Answer Relevancy: {result}") # noqa: T201
elif isinstance(metric, UnBiasedMetric):
score = metric.measure(output)
print(f"Bias Score: {score}") # noqa: T201
elif isinstance(metric, NonToxicMetric):
score = metric.measure(output)
print(f"Toxic Score: {score}") # noqa: T201
else:
raise ValueError(
f"""Metric {metric.__name__} is not supported by deepeval
callbacks."""
)
def on_llm_error(self, error: BaseException, **kwargs: Any) -> None:
"""Do nothing when LLM outputs an error."""
pass
def on_chain_start(
self, serialized: Dict[str, Any], inputs: Dict[str, Any], **kwargs: Any
) -> None:
"""Do nothing when chain starts"""
pass
def on_chain_end(self, outputs: Dict[str, Any], **kwargs: Any) -> None:
"""Do nothing when chain ends."""
pass
def on_chain_error(self, error: BaseException, **kwargs: Any) -> None:
"""Do nothing when LLM chain outputs an error."""
pass
def on_tool_start(
self,
serialized: Dict[str, Any],
input_str: str,
**kwargs: Any,
) -> None:
"""Do nothing when tool starts."""
pass
def on_agent_action(self, action: AgentAction, **kwargs: Any) -> Any:
"""Do nothing when agent takes a specific action."""
pass
def on_tool_end(
self,
output: Any,
observation_prefix: Optional[str] = None,
llm_prefix: Optional[str] = None,
**kwargs: Any,
) -> None:
"""Do nothing when tool ends."""
pass
def on_tool_error(self, error: BaseException, **kwargs: Any) -> None:
"""Do nothing when tool outputs an error."""
pass
def on_text(self, text: str, **kwargs: Any) -> None:
"""Do nothing"""
pass
def on_agent_finish(self, finish: AgentFinish, **kwargs: Any) -> None:
"""Do nothing"""
pass