Hugging Face Hub

Lack of transparency

All models available on Hugging Face inference endpoints are open-weights, but there is not information on the inference infrastructure. Thus, the environmental impacts are estimated with a lower precision.

This guide focuses on the integration of EcoLogits with the Hugging Face Hub official python client .

Official links:

• Repository: huggingface/huggingface_hub
• Documentation: huggingface.co

Installation

To install EcoLogits along with all necessary dependencies for compatibility with the Hugging Face Hub client, please use the huggingface-hub extra-dependency option as follows:

pip install ecologits[huggingface-hub]

This installation command ensures that EcoLogits is set up with the specific libraries required to interface seamlessly with Hugging Face Hub's Python client.

Chat Completions

Example

Integrating EcoLogits with your applications does not alter the standard outputs from the API responses. Instead, it enriches them by adding the Impacts object, which contains detailed environmental impact data.

SyncAsync

from ecologits import EcoLogits
from huggingface_hub import InferenceClient

# Initialize EcoLogits
EcoLogits.init()

client = InferenceClient(model="HuggingFaceH4/zephyr-7b-beta")
response = client.chat_completion(
    messages=[{"role": "user", "content": "Tell me a funny joke!"}],
    max_tokens=15
)

# Get estimated environmental impacts of the inference
print(response.impacts)

import asyncio
from ecologits import EcoLogits
from huggingface_hub import AsyncInferenceClient

# Initialize EcoLogits
EcoLogits.init()

client = AsyncInferenceClient(model="HuggingFaceH4/zephyr-7b-beta")

async def main() -> None:
    response = await client.chat_completion(
        messages=[{"role": "user", "content": "Tell me a funny joke!"}],
        max_tokens=15
    )

    # Get estimated environmental impacts of the inference
    print(response.impacts)


asyncio.run(main())

Streaming example

In streaming mode, the impacts are calculated incrementally, which means you don't need to sum the impacts from each data chunk. Instead, the impact information in the last chunk reflects the total cumulative environmental impacts for the entire request.

SyncAsync

from ecologits import EcoLogits
from huggingface_hub import InferenceClient

# Initialize EcoLogits
EcoLogits.init()

client = InferenceClient(model="HuggingFaceH4/zephyr-7b-beta")
stream = client.chat_completion(
    messages=[{"role": "user", "content": "Tell me a funny joke!"}],
    max_tokens=15,
    stream=True
)

for chunk in stream:
    # Get cumulative estimated environmental impacts of the inference
    print(chunk.impacts)

import asyncio
from ecologits import EcoLogits
from huggingface_hub import AsyncInferenceClient

# Initialize EcoLogits
EcoLogits.init()

client = AsyncInferenceClient(model="HuggingFaceH4/zephyr-7b-beta")

async def main() -> None:
    stream = await client.chat_completion(
        messages=[{"role": "user", "content": "Tell me a funny joke!"}],
        max_tokens=15,
        stream=True
    )

    async for chunk in stream:
        # Get cumulative estimated environmental impacts of the inference
        print(chunk.impacts)


asyncio.run(main())