issues
search
epfLLM
/
Megatron-LLM
distributed trainer for LLMs
Other
529
stars
76
forks
source link
Tokens per second metric
#66
Closed
AleHD
closed
1 year ago
AleHD
commented
1 year ago
Features:
[x] Added tokens/sec metric calculation. The number of tokens is calculated on the fly (and distributed across data parallel groups), so instruction datasets with
--variable_seq_len
will be calculated correctly.
Features:
--variable_seq_len
will be calculated correctly.