LangChain
vLLM 也可通过 LangChain 获取。
运行下面命令安装 LangChain:
pip install langchain langchain_community -q
从 langchain
使用 VLLM
类,在一个或多个 GPU 上运行推理
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 ?"))
请参阅该教程获得更多细节。