mirror of
https://github.com/BerriAI/litellm.git
synced 2025-04-25 02:34:29 +00:00
d57be47b0f
* ci(config.yml): add a 'check_code_quality' step Addresses https://github.com/BerriAI/litellm/issues/5991 * ci(config.yml): check why circle ci doesn't pick up this test * ci(config.yml): fix to run 'check_code_quality' tests * fix(__init__.py): fix unprotected import * fix(__init__.py): don't remove unused imports * build(ruff.toml): update ruff.toml to ignore unused imports * fix: fix: ruff + pyright - fix linting + type-checking errors * fix: fix linting errors * fix(lago.py): fix module init error * fix: fix linting errors * ci(config.yml): cd into correct dir for checks * fix(proxy_server.py): fix linting error * fix(utils.py): fix bare except causes ruff linting errors * fix: ruff - fix remaining linting errors * fix(clickhouse.py): use standard logging object * fix(__init__.py): fix unprotected import * fix: ruff - fix linting errors * fix: fix linting errors * ci(config.yml): cleanup code qa step (formatting handled in local_testing) * fix(_health_endpoints.py): fix ruff linting errors * ci(config.yml): just use ruff in check_code_quality pipeline for now * build(custom_guardrail.py): include missing file * style(embedding_handler.py): fix ruff check
114 lines
3.9 KiB
Python
114 lines
3.9 KiB
Python
"""
|
|
arize AI is OTEL compatible
|
|
|
|
this file has Arize ai specific helper functions
|
|
"""
|
|
|
|
from typing import TYPE_CHECKING, Any, Optional, Union
|
|
|
|
if TYPE_CHECKING:
|
|
from opentelemetry.trace import Span as _Span
|
|
|
|
Span = _Span
|
|
else:
|
|
Span = Any
|
|
|
|
|
|
def set_arize_ai_attributes(span: Span, kwargs, response_obj):
|
|
from litellm.integrations._types.open_inference import (
|
|
MessageAttributes,
|
|
MessageContentAttributes,
|
|
OpenInferenceSpanKindValues,
|
|
SpanAttributes,
|
|
)
|
|
|
|
optional_params = kwargs.get("optional_params", {})
|
|
# litellm_params = kwargs.get("litellm_params", {}) or {}
|
|
|
|
#############################################
|
|
############ LLM CALL METADATA ##############
|
|
#############################################
|
|
# commented out for now - looks like Arize AI could not log this
|
|
# metadata = litellm_params.get("metadata", {}) or {}
|
|
# span.set_attribute(SpanAttributes.METADATA, str(metadata))
|
|
|
|
#############################################
|
|
########## LLM Request Attributes ###########
|
|
#############################################
|
|
|
|
# The name of the LLM a request is being made to
|
|
if kwargs.get("model"):
|
|
span.set_attribute(SpanAttributes.LLM_MODEL_NAME, kwargs.get("model"))
|
|
|
|
span.set_attribute(
|
|
SpanAttributes.OPENINFERENCE_SPAN_KIND, OpenInferenceSpanKindValues.LLM.value
|
|
)
|
|
messages = kwargs.get("messages")
|
|
|
|
# for /chat/completions
|
|
# https://docs.arize.com/arize/large-language-models/tracing/semantic-conventions
|
|
if messages:
|
|
span.set_attribute(
|
|
SpanAttributes.INPUT_VALUE,
|
|
messages[-1].get("content", ""), # get the last message for input
|
|
)
|
|
|
|
# LLM_INPUT_MESSAGES shows up under `input_messages` tab on the span page
|
|
for idx, msg in enumerate(messages):
|
|
# Set the role per message
|
|
span.set_attribute(
|
|
f"{SpanAttributes.LLM_INPUT_MESSAGES}.{idx}.{MessageAttributes.MESSAGE_ROLE}",
|
|
msg["role"],
|
|
)
|
|
# Set the content per message
|
|
span.set_attribute(
|
|
f"{SpanAttributes.LLM_INPUT_MESSAGES}.{idx}.{MessageAttributes.MESSAGE_CONTENT}",
|
|
msg.get("content", ""),
|
|
)
|
|
|
|
# The Generative AI Provider: Azure, OpenAI, etc.
|
|
span.set_attribute(SpanAttributes.LLM_INVOCATION_PARAMETERS, str(optional_params))
|
|
|
|
if optional_params.get("user"):
|
|
span.set_attribute(SpanAttributes.USER_ID, optional_params.get("user"))
|
|
|
|
#############################################
|
|
########## LLM Response Attributes ##########
|
|
# https://docs.arize.com/arize/large-language-models/tracing/semantic-conventions
|
|
#############################################
|
|
for choice in response_obj.get("choices"):
|
|
response_message = choice.get("message", {})
|
|
span.set_attribute(
|
|
SpanAttributes.OUTPUT_VALUE, response_message.get("content", "")
|
|
)
|
|
|
|
# This shows up under `output_messages` tab on the span page
|
|
# This code assumes a single response
|
|
span.set_attribute(
|
|
f"{SpanAttributes.LLM_OUTPUT_MESSAGES}.0.{MessageAttributes.MESSAGE_ROLE}",
|
|
response_message["role"],
|
|
)
|
|
span.set_attribute(
|
|
f"{SpanAttributes.LLM_OUTPUT_MESSAGES}.0.{MessageAttributes.MESSAGE_CONTENT}",
|
|
response_message.get("content", ""),
|
|
)
|
|
|
|
usage = response_obj.get("usage")
|
|
if usage:
|
|
span.set_attribute(
|
|
SpanAttributes.LLM_TOKEN_COUNT_TOTAL,
|
|
usage.get("total_tokens"),
|
|
)
|
|
|
|
# The number of tokens used in the LLM response (completion).
|
|
span.set_attribute(
|
|
SpanAttributes.LLM_TOKEN_COUNT_COMPLETION,
|
|
usage.get("completion_tokens"),
|
|
)
|
|
|
|
# The number of tokens used in the LLM prompt.
|
|
span.set_attribute(
|
|
SpanAttributes.LLM_TOKEN_COUNT_PROMPT,
|
|
usage.get("prompt_tokens"),
|
|
)
|
|
pass
|