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。
限制：数据范围不能太大，否则空间浪费
，详情可参考91视频
Nano Banana 2 延续 Gemini 系列的真实世界知识库，并结合网页搜索实时信息，使模型在空间理解、比例关系、光影处理与中文文本渲染方面表现更自然。，详情可参考旺商聊官方下载
Once the basic stuff was working, I wanted to load TAP files directly, simulating cassette loading. This was the first time the agent missed a few things, specifically about the timing the Spectrum loading routines expected, and here we are in the territory where LLMs start to perform less efficiently: they can’t easily run the SDL emulator and see the border changing as data is received and so forth. I asked Claude Code to do a refactoring so that zx_tick() could be called directly and was not part of zx_frame(), and to make zx_frame() a trivial wrapper. This way it was much simpler to sync EAR with what it expected, without callbacks or the wrong abstractions that it had implemented. After such change, a few minutes later the emulator could load a TAP file emulating the cassette without problems.