muti_gpu_model not supported in crf loss[BUG]

[ZhyiXu]

File "/root/ubuntu/software/anaconda3/lib/python3.7/site-packages/kashgari/utils/crf/crf_losses.py", line 71, in crf_loss
if crf.learn_mode == 'join':
AttributeError: 'Concatenate' object has no attribute 'learn_mode'

[BrikerMan]

@CyberZHG could you help me with the issue. I don't know how to implement a multi-GPU / TPU supported layer.

[CyberZHG]

You don't need to do anything to support multi-GPU. It's because this implementation of CRF layer must be the last layer in 'join' mode. However, a concatenation layer will be created to merge the outputs (from CRF layers) from different GPUs.

[CyberZHG]

I think one way to use this CRF for multi-GPU in 'join' mode is:

• Build the model without the CRF
• Convert to multi-GPU model
• Add CRF layer to the output of the multi-GPU model in CPU context
• Compile the model with the CRF loss