[Unresolved] FastGPT calls nomic-embedding without normalization

The current LLM and embedding model invocation structure is: Vendor/Proxy oneapi fastgpt (and other applications). Neither oneapi nor fastgpt has built-in normalization, causing fastgpt to get results greater than 1 when calling the embedding model to calculate the angular projection distance. The next step is to research API interface forwarding solutions and see whether any ready-made ones are available.