Did the model finish?
Look below your prompt for Response:. Read that text first. On the Qwen2 walkthrough, the answer begins “The capital of France is Paris.” A loaded model with an empty response hasn't answered your question.
Why did the text stop?
max_tokens reached means the output allowance was exhausted. Increase --max-tokens if the sentence is unfinished. An end-of-sequence token means the model chose to stop. Try a second prompt to see both outcomes.
How long did I wait?
The initial download and first build can take longer than answering a question. In the final timing lines, prompt eval covers reading the prompt and decode covers writing the answer. To compare two runs, keep the model, prompt, context length and output limit the same.
See the operations behind an answer
Open the IR Visualizer for this run and compare the prefill and decode kernel sequences.