Training Progress Report Summary | Generated by AI
Healthy and on-track. Quick read:
| Metric | Value | Verdict |
|---|---|---|
| Elapsed | 1h 14m | — |
| Iter | 220 / 8000 (2.75%) | — |
| Per-iter time | 18.53 s (rock-steady) | Better than my 21.4 s estimate |
| Loss trajectory | 10.99 → 9.46 → 7.64 → 7.10 → 6.41 | Healthy descent |
| MFU | 13.7% | Normal for no-compile on 4070 |
| GPU | 100% util, 9.7 GB, 64°C, 212 W | All good |
Revised total ETA: 18.53 s × 8000 = ~41 hours (≈ 1.7 days), faster than my earlier 47h projection. Should finish around Saturday May 16, early morning.
First val checkpoint coming at iter 250 (~9 min from now). That’s the first real quality signal — I’d expect val ≈ 6.0–6.5.
Nothing to do. Check back at iter 250 or iter 500 for the first val loss reading.