langchain/libs/community/langchain_community/callbacks/arize_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

214 lines
7.3 KiB
Python

from datetime import datetime
from typing import Any, Dict, List, Optional
from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.callbacks import BaseCallbackHandler
from langchain_core.outputs import LLMResult
from langchain_community.callbacks.utils import import_pandas
class ArizeCallbackHandler(BaseCallbackHandler):
"""Callback Handler that logs to Arize."""
def __init__(
self,
model_id: Optional[str] = None,
model_version: Optional[str] = None,
SPACE_KEY: Optional[str] = None,
API_KEY: Optional[str] = None,
) -> None:
"""Initialize callback handler."""
super().__init__()
self.model_id = model_id
self.model_version = model_version
self.space_key = SPACE_KEY
self.api_key = API_KEY
self.prompt_records: List[str] = []
self.response_records: List[str] = []
self.prediction_ids: List[str] = []
self.pred_timestamps: List[int] = []
self.response_embeddings: List[float] = []
self.prompt_embeddings: List[float] = []
self.prompt_tokens = 0
self.completion_tokens = 0
self.total_tokens = 0
self.step = 0
from arize.pandas.embeddings import EmbeddingGenerator, UseCases
from arize.pandas.logger import Client
self.generator = EmbeddingGenerator.from_use_case(
use_case=UseCases.NLP.SEQUENCE_CLASSIFICATION,
model_name="distilbert-base-uncased",
tokenizer_max_length=512,
batch_size=256,
)
self.arize_client = Client(space_key=SPACE_KEY, api_key=API_KEY)
if SPACE_KEY == "SPACE_KEY" or API_KEY == "API_KEY":
raise ValueError("❌ CHANGE SPACE AND API KEYS")
else:
print("✅ Arize client setup done! Now you can start using Arize!") # noqa: T201
def on_llm_start(
self, serialized: Dict[str, Any], prompts: List[str], **kwargs: Any
) -> None:
for prompt in prompts:
self.prompt_records.append(prompt.replace("\n", ""))
def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
"""Do nothing."""
pass
def on_llm_end(self, response: LLMResult, **kwargs: Any) -> None:
pd = import_pandas()
from arize.utils.types import (
EmbeddingColumnNames,
Environments,
ModelTypes,
Schema,
)
# Safe check if 'llm_output' and 'token_usage' exist
if response.llm_output and "token_usage" in response.llm_output:
self.prompt_tokens = response.llm_output["token_usage"].get(
"prompt_tokens", 0
)
self.total_tokens = response.llm_output["token_usage"].get(
"total_tokens", 0
)
self.completion_tokens = response.llm_output["token_usage"].get(
"completion_tokens", 0
)
else:
self.prompt_tokens = (
self.total_tokens
) = self.completion_tokens = 0 # assign default value
for generations in response.generations:
for generation in generations:
prompt = self.prompt_records[self.step]
self.step = self.step + 1
prompt_embedding = pd.Series(
self.generator.generate_embeddings(
text_col=pd.Series(prompt.replace("\n", " "))
).reset_index(drop=True)
)
# Assigning text to response_text instead of response
response_text = generation.text.replace("\n", " ")
response_embedding = pd.Series(
self.generator.generate_embeddings(
text_col=pd.Series(generation.text.replace("\n", " "))
).reset_index(drop=True)
)
pred_timestamp = datetime.now().timestamp()
# Define the columns and data
columns = [
"prediction_ts",
"response",
"prompt",
"response_vector",
"prompt_vector",
"prompt_token",
"completion_token",
"total_token",
]
data = [
[
pred_timestamp,
response_text,
prompt,
response_embedding[0],
prompt_embedding[0],
self.prompt_tokens,
self.total_tokens,
self.completion_tokens,
]
]
# Create the DataFrame
df = pd.DataFrame(data, columns=columns)
# Declare prompt and response columns
prompt_columns = EmbeddingColumnNames(
vector_column_name="prompt_vector", data_column_name="prompt"
)
response_columns = EmbeddingColumnNames(
vector_column_name="response_vector", data_column_name="response"
)
schema = Schema(
timestamp_column_name="prediction_ts",
tag_column_names=[
"prompt_token",
"completion_token",
"total_token",
],
prompt_column_names=prompt_columns,
response_column_names=response_columns,
)
response_from_arize = self.arize_client.log(
dataframe=df,
schema=schema,
model_id=self.model_id,
model_version=self.model_version,
model_type=ModelTypes.GENERATIVE_LLM,
environment=Environments.PRODUCTION,
)
if response_from_arize.status_code == 200:
print("✅ Successfully logged data to Arize!") # noqa: T201
else:
print(f'❌ Logging failed "{response_from_arize.text}"') # noqa: T201
def on_llm_error(self, error: BaseException, **kwargs: Any) -> None:
"""Do nothing."""
pass
def on_chain_start(
self, serialized: Dict[str, Any], inputs: Dict[str, Any], **kwargs: Any
) -> None:
pass
def on_chain_end(self, outputs: Dict[str, Any], **kwargs: Any) -> None:
"""Do nothing."""
pass
def on_chain_error(self, error: BaseException, **kwargs: Any) -> None:
"""Do nothing."""
pass
def on_tool_start(
self,
serialized: Dict[str, Any],
input_str: str,
**kwargs: Any,
) -> None:
pass
def on_agent_action(self, action: AgentAction, **kwargs: Any) -> Any:
"""Do nothing."""
pass
def on_tool_end(
self,
output: Any,
observation_prefix: Optional[str] = None,
llm_prefix: Optional[str] = None,
**kwargs: Any,
) -> None:
pass
def on_tool_error(self, error: BaseException, **kwargs: Any) -> None:
pass
def on_text(self, text: str, **kwargs: Any) -> None:
pass
def on_agent_finish(self, finish: AgentFinish, **kwargs: Any) -> None:
pass