Proper count tokens for conversation / message level

[vectorkovacspeter]

Contact Details

kovacs.peter@vector.hu

What is your question?

How can I properly get the input and output token numbers for a conversation or a message?

In the Mongo database, thez messages.tokenCount value is null for the non-user messages, am I search it in the wrong place?

Is there a detailed documentation available for the database schama, and its content?

More Details

I try to sum the tokens used by a specified conversation or a specified message stored in the chat database (I use mongo).

What is the main subject of your question?

Other

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Hi @vectorkovacspeter, thanks for trying out LibreChat.

Are you on the latest commit on main branch? The message.tokenCount is mainly for caching the tokenCount on subsequent message requests.

the Mongo database, thez messages.tokenCount value is null for the non-user messages,

This should not be the case, as you can see below

Is there a detailed documentation available for the database schama, and its content?

No, but you can inspect the schemas in /LibreChat/api/models/schema/

To view what is actually spent, you should be looking at the transactions collection and filter by conversationId

[vectorkovacspeter]

Thanks for your answer, I use the backend and client version: 0.5.8.

As I checked the database again, not all the tokenCount is null, where the sender is ChatGPT, that some of them. It seems its appears on different models, and strop reasons, so I can't find the common thing in them.

I try to update my version, maybe it's solve the problem, but as I customize it som points, its a bit time consuming to update :)

[danny-avila]

See these docs and PR for more info:

https://docs.librechat.ai/features/token_usage.html

#1018