Training Progress Update | Generated by AI

Home 2026.06

Got it — letting it run. Here’s the status:

Progress:   400 / 2,656 steps (15%)
Elapsed:    ~2 hours
ETA:        ~9 more hours (around 04:00 tomorrow)
Loss:       2.86 → 1.70 (healthy)
GPU:        100% utilization, 11.5GB

Checkpoints at lzw-notes-lora/checkpoint-{10,200,400}/. Final adapter will save to lzw-notes-lora/. You can check progress anytime with:

cat /mnt/data/zz/finetune/lzw-notes-lora/checkpoint-*/trainer_state.json | python3.11 -c "
import json, sys
for line in sys.stdin:
    try:
        state = json.loads(line)
        print(f'Step {state[\"global_step\"]}/{state[\"max_steps\"]} epoch={state[\"epoch\"]:.2f}')
    except: pass
" | tail -1
Back Donate