bofeng huang

Just realized that the modified `_validate_prompt_and_tokenize` function is used for completion serving, chat serving, and embedding serving. However, we only need to disable `add_special_tokens` for chat serving, as these special...

> Whether openai.types.CompletionUsage.completion_tokens == openai.types.CompletionUsage.prompt_tokens in general, which would be the case if BOS token is not added by default. Not sure I fully understand. Why should we compare input...

@DarkLight1337 thanks for the discussion!