Reuse the same model
Keep the model and context length the same. Change the prompt to a task where you can judge the answer yourself.
version/v8/scripts/cks-v8-run run \
hf://Qwen/Qwen2-0.5B-Instruct-GGUF/qwen2-0_5b-instruct-q4_k_m.gguf \
--context-len 1024 \
--prompt 'List three vegetables, one per line.' \
--max-tokens 48 \
--temperature 0.0Our run returned carrots, cabbage and lettuce on separate numbered lines, then printed stop: eos token 151645. EOS means the model ended its response before using the 48-token allowance.
Check two things separately. Did the runner produce text? Did the model actually follow your request for three lines? A small model can run successfully and still give a poor answer.
When an answer cuts off
Repeat the command with --max-tokens 96 if an answer ends mid-sentence at the output limit. A larger cap allows a longer answer. It doesn’t force the model to use every token.
Choose your next task
For an interactive session, omit the entire --prompt option and its quoted text. For a different model, start with an exact artifact listed in the v8 runbook. A larger model can require much more memory. Speech transcription also has its own command and takes an audio file as input.