OpenTelemetry support

With the support of OpenTelemetry in EcoLogits you can monitor request metrics such as the number of input/output tokens or the environmental impacts in time.

Installation

To install EcoLogits along with all necessary dependencies for compatibility with OpenTelemetry, please use the opentelemetry extra-dependency option as follows:

pip install ecologits[opentelemetry] # (1)!

1. Also include the provider you use in the installation command.

This installation command ensures that EcoLogits is set up with the specific libraries required to work seamlessly with OpenTelemetry.

Configuration

To configure OpenTelemetry with EcoLogits, you will need to provide the URL endpoint at the initialization.

from ecologits import EcoLogits
from openai import OpenAI

# Configure OpenTelemetry endpoint in EcoLogits
EcoLogits.init(
    providers=["openai"],
    opentelemetry_endpoint='https://localhost:4318/v1/metrics'
)

client = OpenAI()

# All requests will log ecologits metrics
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Tell me a funny joke!"}]
)

Metrics

EcoLogits will automatically report the following metrics when OpenTelemetry is enabled.

Different units for impact metrics

Notice that the units used for impacts differ from the original ImpactsOutput object, in order to respect OpenTelemetry guidelines.

Metric name	Type	Description
ecologits_requests_total	Counter	Number of requests
ecologits_input_tokens_ratio	Gauge	Number of input tokens
ecologits_output_tokens_ratio	Gauge	Number of output tokens
ecologits_request_latency_seconds	Gauge	Request latency in seconds
ecologits_energy_joules	Gauge	Average energy consumption of the request in joules
ecologits_gwp_gCO2eq	Gauge	Average GWP impacts of the request in gCO2eq
ecologits_adpe_gSbeq	Gauge	Average ADPe impacts of the request in gSbeq
ecologits_pe_joules	Gauge	Average PE impacts of the request in joules

Labels

By default, EcoLogits will report three labels:

• provider: Name of the provider
• model: Name of the model used
• endpoint: API endpoint of the request

Configure custom labels

It is possible to add custom labels to any request using the EcoLogits.label method.

With a context manager

from ecologits import EcoLogits
from openai import OpenAI

EcoLogits.init(
    providers=["openai"], 
    opentelemetry_endpoint="http://localhost:4318/v1/metrics"
)
client = OpenAI()

with EcoLogits.label(my_custom_label="test"): # (1)! 
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "Tell me a funny joke"}]
    )

1. Also works with asynchronous code with async with syntax.

With a function decorator

from ecologits import EcoLogits
from openai import OpenAI

EcoLogits.init(
    providers=["openai"], 
    opentelemetry_endpoint="http://localhost:4318/v1/metrics"
)
client = OpenAI()

@EcoLogits.label(task="summarization")  # (1)! 
def summarize(text: str) -> str:
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": f"Summarize this:\n\n{text}"}]
    )
    return response.choices[0].message.content

1. Also works with asynchronous functions with async def syntax.