Update BeanOutputConverter.java

[chang6666]

Resolved the exceptional issue of converting the large model deployed locally with Ollama into a Java Bean.

[markpollack]

I think we are going to have to opt into this 'exclusion' in the output converter. I'd like to make this a pluggable stragtegy, who knows how many ways the thinking response will be done. is this from anthropic or deepseek?. we have to overhaul the whole code base to handle 'thinking' responses in the assistant messgae, but hacks like this are useful.

also missing semicolon

[YunKuiLu]

Should it be \s and \S here?

Suggested change

	text = text.replaceAll("<think>[\\s\\s]*?</think>","")
	text = text.replaceAll("<think>[\\s\\S]*?</think>","");

Although this regex is simple, it's better to have unit tests.

[nicolaskrier]

I have implemented something similar for qwen3:14b model running locally on my Ollama. The solution is a bit more complex but it works properly until now.

@Override
public T convert(@NonNull String text) {
    var thinkingPattern = Pattern.compile("^<think>[\\w\\s\\p{P}]*</think>", Pattern.MULTILINE);
    var thinkingMatcher = thinkingPattern.matcher(text);
    var cleanedText = thinkingMatcher.replaceFirst("");

    return super.convert(cleanedText);
}