- 训练代码
from keras.utils import multi_gpu_model # 0 < gpus <= 你真实含有的gpu数量 parallel_model = multi_gpu_model(model, gpus=8) # 进行编译 parallel_model.compile(loss='categorical_crossentropy', optimizer='rmsprop') # fit 方法会在8个gpu上分布运行，就是将数据一个batch里面的数据平均分到8个gpu上 # 由于batch_size = 256,每块gpu有256/8数据在运行 parallel_model.fit(x, y, epochs=20, batch_size=256) - 问题
- ValueError: You are trying to load a weight file containing 3 layers into a model with 1 layers.
#加载模型的时候，先将模型创建起来 model = create_model() model = multi_gpu_model(model , gpus = 8) - Check failed: work_element_count > 0 (0 vs. 0)
tensorflow==1.8.0 -> tensorflow==1.7.0
- ValueError: You are trying to load a weight file containing 3 layers into a model with 1 layers.
- 参考资料
