deepglint / deepglint/EasyQuant
TensorRT统计blob直方图
Open
- Dominant language
- Python
- Stars
- 407
- Forks
- 67
- PR merge metrics
- No merged PRs in 30d
Description
您好,在TensorRT中,我的理解是[-max, max] 量化为[-127, 127].
可在TensorRT激活值量化部分,也就是tools/caffe_quanttable_e2e.py收集直方图的部分:
print("\nCollect histograms of activations:")
for i, image in enumerate(images_files):
net_forward(net, image, transformer)
for layer in quantize_layer_lists:
blob = net.blobs[layer.blob_name].data[0].flatten()
blob= blob[blob > 0]
layer.initial_histograms(blob)
if i % 100 == 0:
print("loop stage 2 : %d/%d" % (i, len(images_files)))
为什么只收集blob>0的部分呢,如果卷积后面没有ReLU的情况怎么处理呢?
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.