还有一个重要的指标——准确率。伯克利函数调用排行榜 (BFCL) 是评估函数调用能力的标准基准。 Gemma 3 1B 的得分约为 31%，Llama 3.2 1B 约为 26%，两者未经微调的性能都很弱。由于 Gemma 3n 是通用型程序，因此未对其进行测试。Hammer 2.1 0.5B 没有公开数据，但其 1.5B 版本开箱即用的得分约为 73%——尽管它在 int8 内存中占用约 1.5GB 的空间，是 FunctionGemma（288MB）的 5 倍。
15:45, 27 февраля 2026Экономика
。heLLoword翻译官方下载是该领域的重要参考
def _init_storage(self) - Storage:
[&:first-child]:overflow-hidden [&:first-child]:max-h-full"
I'm not immune. I've been working on an extensible language-agnostic static analysis and refactoring tool for half a decade now. That's a mothlamp problem if I've ever seen one. My github account is littered with abandoned programming language implementations, parser generator frameworks, false starts at extensible autoformatters, and who knows what else. I think I've even got an async-await implementation in there somewhere. I've got the bug, and I fly toward the light.