Token Usage is wrong. Completion Token counted twice

[niklasfink]

I've recently tested the LangChain token usage, which unfortunately doesn't work yet with streaming models, but while testing, I saw that the self-coded calculation in gpt-engineer probably has an error around here:

gpt-engineer/gpt_engineer/ai.py

Line 174 in eebbe1b

messages.append(response)

Due to this line, the following calculation error occurs. Example:
Prompt Token Usage: 1000
Completion Token Usage: 500

gpt-engineer would calculate 1500 as Prompt Token, 500 as Completion Token and 2000 Total.

[niklasfink]

response = self.llm(messages, callbacks=callsbacks)  # type: ignore

self.update_token_usage_log(messages=messages, answer=response.content, step_name=step_name)

messages.append(response)
logger.debug(f"Chat completion finished: {messages}")

return messages

This would fix it. Calculation is only off by -1 on gpt-engineer prompt side then.