You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Dear author, regarding the issue of training speed #37
From my experience, when training a Qwen-2.5-3B model on 8×8 A100 (40G) nodes with a batch size of 256, the per-step time ranges from 40 to 50 minutes. Is this speed normal?
From my experience, when training a Qwen-2.5-3B model on 8×8 A100 (40G) nodes with a batch size of 256, the per-step time ranges from 40 to 50 minutes. Is this speed normal?