pytorch-quantization-demo

pytorch-quantization-demo copied to clipboard

split freeze and fix some small bugs

1

1. 将`freeze()`拆成`reuse_qparam()`和`freeze()`，前者复用`qparam`，后者负责提前计算。这样做的优点是每次推理的时候免除了`freeze()`部分的计算（例如`M`），缺点是部分`qparam`被重复保存。 2. 统一`buffer`被保存的`shape`为`torch.Size([])`，避免`M`读取时的类型冲突。 3. 添加了一些操作，非常粗糙地模拟32位定点数乘法。 4. 修复了`QReLU`和`QMaxPooling2d`会多存一个`qo`的bug。

crazy-JiangDongHua

FileNotFoundError: [Errno 2] No such file or directory: 'ckpt/mnist_cnnbn.pt'

1

when i execute `python quantization_aware_training.py`, it outputs the error that: ```bash Traceback (most recent call last): File "quantization_aware_training.py", line 81, in model.load_state_dict(torch.load('ckpt/mnist_cnnbn.pt', map_location='cpu')) File "/home/jyg/anaconda3/envs/mixcompre/lib/python3.8/site-packages/torch/serialization.py", line 791, in load with...

yigengjiang