mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2025-12-14 18:25:01 +08:00
31 lines
773 B
Markdown
31 lines
773 B
Markdown
(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.
|