例如，微软graphrag 的 settings.yaml emddings部分设成：
...
embeddings:
## parallelization: override the global parallelization settings for embeddings
async_mode: threaded # or asyncio
llm:
api_key: 你的HF_TOKEN
type: openai_embedding # or azure_openai_embedding
model: any
api_base: https://mikeee-emb384-oai.hf.space/v1
...
不过跑中文还是很费 llm 令牌，openrouter 上 gpt-4o-mini（比3.5-turbo都便宜，4o-mini $0.15 $0.6，3.5-turbo $0.5 $1.5）5刀全部用光了红楼梦还没跑完