Openai Completion Client For Lora

Refer to the trtllm-serve documentation for starting a server.

Source NVIDIA/TensorRT-LLM.

### OpenAI Completion Client

import os
from pathlib import Path

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="tensorrt_llm",
)

lora_path = Path(
    os.environ.get("LLM_MODELS_ROOT")) / "llama-models" / "luotuo-lora-7b-0.1"
assert lora_path.exists(), f"Lora path {lora_path} does not exist"

response = client.completions.create(
    model="llama-7b-hf",
    prompt="美国的首都在哪里? \n答案:",
    max_tokens=20,
    extra_body={
        "lora_request": {
            "lora_name": "luotuo-lora-7b-0.1",
            "lora_int_id": 0,
            "lora_path": str(lora_path)
        }
    },
)

print(response)