LangChain

vLLM is also available via LangChain .

To install LangChain, run

pip install langchain langchain_community -q

To run inference on a single or multiple GPUs, use VLLM class from langchain.

Code

from langchain_community.llms import VLLM

llm = VLLM(model="mosaicml/mpt-7b",
        trust_remote_code=True,  # mandatory for hf models
        max_new_tokens=128,
        top_k=10,
        top_p=0.95,
        temperature=0.8,
        # tensor_parallel_size=... # for distributed inference
)

print(llm("What is the capital of France ?"))

Please refer to this Tutorial for more details.