How does it calculate the "% context left" ?

[yashasvigirdhar]

I am just curious about this. How does the cli calculates this and if I can get more information into this, for e.g. the absolute no of tokens left.

[stanton182]

Have to agree, and also if it is time barred basis context window per 1 minute then a countdown or something may be more appropriate

[scidomino]

The context is basically the chat history. Whenever you talk to GCLI or it talks to you, or it calls a function or gets a function response, all that is the context. That's what we send to the server when we make an inference call. The actual context limit depends on which model you are using. For most models its a bit over a million tokens (but tokens don't correspond to words or letters so don't bother trying to figure out precisely what your token count ought to be).

If you are close to filling up your context, you can try running the /compress command which basically replaces your context with a LLM generated summary that's smaller. That said, do this automatically whenever you exceed 70% of your context window at the end of a turn.

Note the context size does not depend on time. We do not sunset anything automatically.

[yashasvigirdhar]

Thanks! This helps.