为什么都是minicpm，一个是40层decoder，一个却是52层

[CHNtentes]

https://huggingface.co/openbmb/MiniCPM-2B-128k 这里下载的模型是40层，但是安卓apk安装的模型却显示有52层

./ndarray-cache.json: "name": "model.layers.51.input_layernorm.weight", ./ndarray-cache.json: "name": "model.layers.51.self_attn.qkv_proj.q_weight", ./ndarray-cache.json: "name": "model.layers.51.self_attn.qkv_proj.q_scale", ./ndarray-cache.json: "name": "model.layers.51.self_attn.qkv_proj.q_zero", ./ndarray-cache.json: "name": "model.layers.51.self_attn.o_proj.q_weight", ./ndarray-cache.json: "name": "model.layers.51.self_attn.o_proj.q_scale", ./ndarray-cache.json: "name": "model.layers.51.self_attn.o_proj.q_zero", ./ndarray-cache.json: "name": "model.layers.51.post_attention_layernorm.weight", ./ndarray-cache.json: "name": "model.layers.51.mlp.gate_up_proj.q_weight", ./ndarray-cache.json: "name": "model.layers.51.mlp.gate_up_proj.q_scale", ./ndarray-cache.json: "name": "model.layers.51.mlp.gate_up_proj.q_zero", ./ndarray-cache.json: "name": "model.layers.51.mlp.down_proj.q_weight", ./ndarray-cache.json: "name": "model.layers.51.mlp.down_proj.q_scale", ./ndarray-cache.json: "name": "model.layers.51.mlp.down_proj.q_zero",

[Achazwl]

APK 2.0 corresponds to MiniCPM-1B, while APK 1.0 corresponds to MiniCPM-2B (not 128k version)