train/mfu metric incorrectly zero | Generated by AI

Home 2026.06

MLflow DB is healthy. Current state:

2 runs in the database:
  1. RUNNING  (started Jun 10 ~20:37 UTC)
  2. FINISHED (started Jun 10 ~14:31 UTC)

Latest metrics at step 700:
  train/loss       = 3.707
  train/tok_per_sec = 55,780
  train/mfu        = 0.0        <-- this is wrong, should be non-zero
  train/dt         = 1.175s per step
  total_train_time = 810s (~13.5 min)

The DB is live and updating. The RUNNING run is actively logging. What’s the issue you’re seeing? Is it:

  1. The UI can’t connect to this DB?
  2. The train/mfu = 0.0 looks wrong?
  3. Something else?
Back Donate