XGZhang

Results 16 comments of XGZhang

可以采用https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/contrib/slim/quantization/post_training_quantization.py#L1064 将模型权重保存为int8格式,并将权重的scale保存为该op的属性。 在推理端加载时,在load_combine这个op中将模型权重再转换为fp32,然后使用scale将权重反量化。

Sorry for not replying in time. It seems that there is something wrong with ShapeInference. Can you provide more information about the code and model you use?

抱歉未能及时回复,能否提供更多paddle相关代码库的版本信息

shape对不上,有可能是输入数据的问题

是不是数据读取不正常,可以检查一下吗

请问是否能提供下所用的代码和版本信息