Warning when an input prompt was truncated in Ollama

[agatheblues]

Hello!

Thanks for the great library. I am trying to find out what's a good way to warn the developer when Ollama truncated their input prompt. We are currently at a time where we want to evaluate whether our prompts are efficient, and not knowing that the prompt was truncated seems like an oversight!

So far the most reliable way seems to be running ollama with ollama serve and look for logs like:

time=2025-09-11T09:35:33.169+02:00 level=WARN source=runner.go:128 msg="truncating input prompt" limit=36 prompt=2443 keep=5 new=36

But that feels quite easy to miss. Of course I can increase the num_ctx to a number large enough that it would be very unlikely that we reach the limit but I am still interested in how one would solve the problem :)

Some ideas:

• The warning could be a clear, red, error log in the server - (what matters for me is that the developer would know straight up that they can't use this chat to evaluate anything)
• Like other implementation I've seen, just error on chat message generation if the limit is hit
• Using the prompt_eval_count that is return with the message? If the metric could be compared to whatever num_ctx was supplied?

Am I missing something obvious? Or maybe this is actually very hard to solve and looking at ollama's logs is still the best approach. Let me know what you think!