Run Inference on HUGS

HUGS, as already mentioned, is based on Text Generation Inference (TGI); meaning that running inference over a deployed HUGS container is exactly the same as for TGI. For more information, please refer to Text Generation Inference Documentation on how to consume TGI.

In the inference examples show below, the host is assumed to be localhost which is the case when deploying HUGS via Kubernetes with port-forwarding or when deploying it with docker run on the current instance. If you have deployed HUGS on Kubernetes using an ingress under a specific IP, host, and/or with SSL (HTTPS), note that you should update the localhost references below with your host or IP.

Messages API

The Messages API is an OpenAI-compatible endpoint under /v1/chat/completions following the OpenAI OpenAPI Specification. Being OpenAI-compatible implies that the inference can be run not only with cURL but also with both the huggingface_hub.InferenceClient and the openai.OpenAI SDK in Python, as well as any other OpenAI-compatible SDK in any programming language.

cURL

Using cURL is pretty straight forward to install and use.

curl http://localhost:8080/v1/chat/completions \
    -X POST \
    -d '{"messages":[{"role":"user","content":"What is Deep Learning?"}],"temperature":0.7,"top_p":0.95,"max_tokens":128}}' \
    -H 'Content-Type: application/json'

Python

As already mentioned, you can either use the huggingface_hub.InferenceClient from the huggingface_hub Python SDK (recommended), the openai Python SDK, or any SDK with an OpenAI-compatible interface that can consume the Messages API.

huggingface_hub

You can install it via pip as pip install --upgrade --quiet huggingface_hub, and then run the following snippet to mimic the cURL commands above i.e. sending requests to the Messages API:

from huggingface_hub import InferenceClient

client = InferenceClient(base_url="http://localhost:8080", api_key="-")

chat_completion = client.chat.completions.create(
    messages=[
        {"role":"user","content":"What is Deep Learning?"},
    ],
    temperature=0.7,
    top_p=0.95,
    max_tokens=128,
)

openai

Alternatively, you can also use the Messages API via openai; you can install it via pip as pip install --upgrade openai, and then run:

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8080/v1/", api_key="-")

chat_completion = client.chat.completions.create(
    model="tgi",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is Deep Learning?"},
    ],
    temperature=0.7,
    top_p=0.95,
    max_tokens=128,
)

Other endpoints

Besides the endpoints mentioned above, TGI also comes with other endpoints as defined in the TGI OpenAPI Specification that can be used not also for inference but also for tokenization, metrics, or information about the deployed model.

< > Update on GitHub