Limit token usage output parameter across all queried llm models

I see hugging face implenentation asserts them but other models like gpt does not.

When the embeddings is past to say openai. Openai has a token output limiting control. Surely this should translate to these api's?