3 ms·

note that the fairseq implementation requires 2x 32GB v100 gpus for inference.

• visarga 6y ago
  And that's why we won't see this deployed in production anywhere.