[server] Update OpenAI Model Support

[dsikka]

Summary

• This PR is focused on updating the openai integration such that it is more independent of the deepsparse server, leveraging the server refactor
• This PR allows multiple text generation models to hosted using the single /v1/chat/completions endpoint, adds the /v1/models endpoint, updates the base routes, and also adds a separate openai specific workflow command

Examples and testing

• On the command line, the following command can be used:

deepsparse.openai sample_config.yaml 

Here, sample_config.yaml has the following structure:

num_cores: 2
num_workers: 2
endpoints:
  - task: text_generation
    model: zoo:nlg/text_generation/opt-1.3b/pytorch/huggingface/opt_pretrain/pruned50_quantW8A8-none

• This launches a FASTAPI app with the following endpoints:

Use the openai api to send requests:

import openai


openai.api_key = "EMPTY"
openai.api_base = "http://localhost:5543/v1"

# Completion API
stream = False
completion = openai.ChatCompletion.create(
    messages="how are you?",
    stream=stream,
    max_tokens=30,
    model="zoo:nlg/text_generation/opt-1.3b/pytorch/huggingface/opt_pretrain/pruned50_quantW8A8-none",
)

print("Chat results:")
if stream:
    text = ""
    for c in completion:
        print(c)
else:
    print(completion)

• The user can provide the model they want to run inference on as part of the payload. If that model was not provided when launching the server, an error will be returned. We will be adding functionality to allow the user to add additional models, after the server has already been launched.

[mgoin]

Really nice! Just two nits