Megatron-LM
Megatron-LM copied to clipboard
[QUESTION] Hello, a consumed samples means how many token in the training? And json file convert to .bin and .idx file
Hello, (1)A consumed samples means how many token in the training? (2)How compute all token number after json file convert to .bin and .idx file?
Marking as stale. No activity in 60 days.
(1) tokens = seq_len * consumed samples
Marking as stale. No activity in 60 days.