Training Progress Snapshot | Generated by AI
Training is running smoothly in the train tmux session. Here’s the snapshot:
Iteration: ~326,960 Latest loss: 2.1948 (iter 326960) Step time: ~621-622ms (very stable) MFU: 14.44% (plateaued)
The MFU (model FLOPs utilization) climbed from ~13.9% to ~14.44% early on and has been flat there for a while. Loss is fluctuating between ~1.5 and ~4.0, which is normal for per-step loss — the running average matters more. Step time is rock solid at 621-622ms, no slowdowns or stalls.