Error while loading the pre-trained models during RL training

Hello Chen,

Thanks for providing this code. It seems really helpful for my current research.

However, I am having issues with making this code work. I have setup the environment as suggested but still able to load the pre-trained models (both extractive and abstractive models).
1. Extractive Model gives error in line `assert ext_meta['net'] == 'ml_rnn_extractor'`. Looking at the `meta.json` file, `net:rnn-ext_abs_rl`. I am not sure why is this discrepancy.
2. For abstractive model, I face `RuntimeError: CUDNN_STATUS_EXECUTION_FAILED` error in line `self._net = abstractor.to(self._device)`. I am not sure how to solve this error. I made sure that CUDA is available. Also, it is not the OOM memory as suggested in some of the pages since the GPU memory never exceeds 1 GB.

It would be great if you could help me out.

Thanks,
Naman

Start with the pretrained-model loading paths at the extractive assertion and at self._net = abstractor.to(self._device), then compare the expected network name with meta.json and investigate the reported CUDNN_STATUS_EXECUTION_FAILED error. Done means both extractive and abstractive pretrained models load successfully and RL training can proceed.

Written by the indexing model from the issue text.