langchain.md

(serving-langchain)=

# LangChain

vLLM is also available via [LangChain](https://github.com/langchain-ai/langchain) .

To install LangChain, run

```console
$ pip install langchain langchain_community -q
```

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

```python
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](https://python.langchain.com/docs/integrations/llms/vllm) for more details.