maxim-py 1.5.6

Maxim Python Library

Maxim SDK

This is Python SDK for enabling Maxim observability. Maxim is an enterprise grade evaluation and observability platform.

How to integrate

Install

pip install maxim-py

Initialize Maxim logger

from maxim.maxim import Maxim

maxim = Maxim(Config(apiKey=apiKey))

Start sending traces

from maxim.logger.logger import LoggerConfig
# Initializing logger
logger = maxim.logger(LoggerConfig(id="log-repository-id"))
# Initializing a new trace
trace = logger.trace(TraceConfig(id="trace-id",name="trace-name",tags={"key":"value"}))
# Creating the generation
generation = trace.generation(GenerationConfig(id=str(uuid4()), model="text-davinci-002", provider="azure", model_parameters={"temperature": 0.7, "max_tokens": 100}))
# Making LLM call
completion = self.client.completions.create(
   model="text-davinci-002",
   prompt="Translate the following English text to French: 'Hello, how are you?'",
   max_tokens=100,
   temperature=0.7
)
# Updating generation
generation.result(completion)
# Ending trace
trace.end()

Integrations with other frameworks

Langchain

We have built in Langchain tracer support

logger = self.maxim.logger(LoggerConfig(id=repoId))
trace_id = str(uuid4())
trace = logger.trace(TraceConfig(
   id=trace_id, name="pre-defined-trace"))

model = OpenAI(callbacks=[MaximLangchainTracer(logger)],api_key=openAIKey)
messages = [
   (
         "system",
         "You are a helpful assistant that translates English to French. Translate the user sentence.",
   ),
   ("human", "I love programming."),
]
model.invoke(messages, config={
   "metadata": {
         "maxim": {
            "trace_id": trace_id,
            "generation_name": "get-answer",
            "generation_tags": {
               "test": "123"
            }
         }
   }
})
trace.event(id=str(uuid4()), name="test event")
trace.end()

Langchain module compatibility

	Anthropic	Bedrock Anthropic	Bedrock Meta	OpenAI	Azure
Chat	✅	✅	✅	✅	✅
Tool call	✅	✅	❓	✅	✅
Chain (via LLM)	✅	✅	✅	✅	✅
Streaming	✅	✅	✅	✅	✳️ Token usage is not supported by Langchain
Agent	⛔️	⛔️	⛔️	⛔️	⛔️

Please reach out to us if you need support for any other package + provider + classes.

Version changelog

v1.5.6

• Improvement - Improved langchain support

v1.5.5

• Improvement - Improves cleanups for log writer for quick returns.

v1.5.4

• Improvement - Improved fs access checks.
• Improvement - Fixes threading locks for periodic syncs in Python3.9

v1.5.3

• Improvement - Adds lambda env support for SDK with no access to filesystem.

v1.5.2

• Feature - Adds support to new langchain_openai.AzureChatOpenAI class in langchain tracer

v1.5.1

• Fix - Adds Python 3.9 compatibility

v1.5.0

• Improvement - Updates connection pool to use session that enforces re-connects before making API calls.

v1.4.5

• Improvement - Adds backoff retries to failed REST calls.

v1.4.4

• Improvement - langchain becomes optional dependency

v1.4.3

• Fix - connection pooling for network calls.
• Fix - connection close issue.

v1.4.2 (🚧 Yanked)

• Fix - connection close issue

v1.4.1

• Adds validation for provider in generation

v1.4.0

• Now generation.result accepts
  • OpenAI chat completion object
  • Azure OpenAI chat completion object
  • Langchain LLMResult, AIMessage object

v1.3.4

• Fixes message_parser

v1.3.2

• Fixes utility function for langchain to parse AIMessage into Maxim logger completion result

v1.3.1

• Adds tool call parsing support for Langchain tracer

v1.3.0

• Adds support for ChatCompletion in generations
• Adds type safety for retrieval results

v1.2.7

• Bug fix where input sent with trace.config was getting overridden with None

v1.2.6

• Adds trace.set_input and trace.set_output methods to control what to show in logs dashboard

v1.2.5

• Removes one no_op command while creating spans
• Minor bug fixes

v1.2.1

• Fixed MaximLangchainTracer error logging flow.

v1.2.0

• Adds langchain support
• Adds local parsers to validate payloads on client side

v1.1.0

• Minor bug fixes around log writer cleanup

v1.0.0

• Public release