Created by: suchenzang
Last bit of https://github.com/facebookresearch/metaseq/pull/197
Removed:
- cuda_utils.cu file - was created with the lightweight conv stuff back in the day: https://github.com/fairinternal/fairseq-py/pull/547
- FeedForwardNetwork class - all calls are made directly to _ffn method instead
- transformer_lm_gpt2_bigger conf - will remove more of these confs later when fixing our configuration setup