wandb:                 eval/f1 0.87565
wandb:               eval/loss 0.70447
wandb:                  eval/p 0.8745
wandb:                  eval/r 0.87847

Note:

1. [evidence_text][SEP][claim]
2. Only trained/validated on instances length <= 512 tokens.