Why jacobi_sample_multilevel() fill window with argmax instead of also using sampling?

[yangbohust]

I have a question, Please ask the great author to help answer this question.

Why jacobi_sample_multilevel() fill window with argmax instead of also using sampling?

code location:

LookaheadDecoding/lade/decoding.py

Line 466 in 9d50de4

past_tokens[1] = torch.argmax(outputs.inp_logits, dim=-1)[0].tolist() #fill window with argmax

LookaheadDecoding/lade/decoding.py

Line 478 in 9d50de4

past_tokens[fill_level + 1] = torch.argmax(outputs.inp_logits, dim=-1)[0].tolist()[1:] #fill window with argmax

Looking forward to your reply.
Thanks