将模型的权重量化以后怎么加载

[pc123s]

采用int4量化了模型，但是量化以后无法打印原始模型的权重了
(music-generate) (base) pengchuang@287defcbcc36:~/music_generate/rwkv.cpp-master$ /home/pengchuang/miniforge3/envs/music-generate/bin/python /home/pengchuang/music_generate/rwkv.cpp-master/python/test.py
Version: 101, Vocabulary size: 20096, Embedding size: 512, Layers: 29, FP16: True
Loaded emb.weight: shape [20096, 512], dtype <class 'numpy.float16'>
Loaded blocks.0.ln1.weight: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.ln1.bias: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.ln2.weight: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.ln2.bias: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.ln0.weight: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.ln0.bias: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.att.time_decay: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.att.time_first: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.att.time_mix_k: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.att.time_mix_v: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.att.time_mix_r: shape [512], dtype <class 'numpy.float32'>
Loaded blocks.0.att.key.weight: shape [512, 512], dtype <class 'numpy.float16'>